2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir_sf.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_dmapi.h"
40 #include "xfs_mount.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_itable.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_alloc.h"
47 #include "xfs_rtalloc.h"
48 #include "xfs_error.h"
49 #include "xfs_dir_leaf.h"
50 #include "xfs_attr_leaf.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.h"
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
62 kmem_zone_t *xfs_bmap_free_item_zone;
65 * Prototypes for internal bmap routines.
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
72 STATIC int /* error */
73 xfs_bmap_add_attrfork_extents(
74 xfs_trans_t *tp, /* transaction pointer */
75 xfs_inode_t *ip, /* incore inode pointer */
76 xfs_fsblock_t *firstblock, /* first block allocated */
77 xfs_bmap_free_t *flist, /* blocks to free at commit */
78 int *flags); /* inode logging flags */
81 * Called from xfs_bmap_add_attrfork to handle local format files.
83 STATIC int /* error */
84 xfs_bmap_add_attrfork_local(
85 xfs_trans_t *tp, /* transaction pointer */
86 xfs_inode_t *ip, /* incore inode pointer */
87 xfs_fsblock_t *firstblock, /* first block allocated */
88 xfs_bmap_free_t *flist, /* blocks to free at commit */
89 int *flags); /* inode logging flags */
92 * Called by xfs_bmapi to update file extent records and the btree
93 * after allocating space (or doing a delayed allocation).
95 STATIC int /* error */
97 xfs_inode_t *ip, /* incore inode pointer */
98 xfs_extnum_t idx, /* extent number to update/insert */
99 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
100 xfs_bmbt_irec_t *new, /* new data to add to file extents */
101 xfs_fsblock_t *first, /* pointer to firstblock variable */
102 xfs_bmap_free_t *flist, /* list of extents to be freed */
103 int *logflagsp, /* inode logging flags */
104 xfs_extdelta_t *delta, /* Change made to incore extents */
105 int whichfork, /* data or attr fork */
106 int rsvd); /* OK to allocate reserved blocks */
109 * Called by xfs_bmap_add_extent to handle cases converting a delayed
110 * allocation to a real allocation.
112 STATIC int /* error */
113 xfs_bmap_add_extent_delay_real(
114 xfs_inode_t *ip, /* incore inode pointer */
115 xfs_extnum_t idx, /* extent number to update/insert */
116 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
117 xfs_bmbt_irec_t *new, /* new data to add to file extents */
118 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
119 xfs_fsblock_t *first, /* pointer to firstblock variable */
120 xfs_bmap_free_t *flist, /* list of extents to be freed */
121 int *logflagsp, /* inode logging flags */
122 xfs_extdelta_t *delta, /* Change made to incore extents */
123 int rsvd); /* OK to allocate reserved blocks */
126 * Called by xfs_bmap_add_extent to handle cases converting a hole
127 * to a delayed allocation.
129 STATIC int /* error */
130 xfs_bmap_add_extent_hole_delay(
131 xfs_inode_t *ip, /* incore inode pointer */
132 xfs_extnum_t idx, /* extent number to update/insert */
133 xfs_btree_cur_t *cur, /* if null, not a btree */
134 xfs_bmbt_irec_t *new, /* new data to add to file extents */
135 int *logflagsp,/* inode logging flags */
136 xfs_extdelta_t *delta, /* Change made to incore extents */
137 int rsvd); /* OK to allocate reserved blocks */
140 * Called by xfs_bmap_add_extent to handle cases converting a hole
141 * to a real allocation.
143 STATIC int /* error */
144 xfs_bmap_add_extent_hole_real(
145 xfs_inode_t *ip, /* incore inode pointer */
146 xfs_extnum_t idx, /* extent number to update/insert */
147 xfs_btree_cur_t *cur, /* if null, not a btree */
148 xfs_bmbt_irec_t *new, /* new data to add to file extents */
149 int *logflagsp, /* inode logging flags */
150 xfs_extdelta_t *delta, /* Change made to incore extents */
151 int whichfork); /* data or attr fork */
154 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
155 * allocation to a real allocation or vice versa.
157 STATIC int /* error */
158 xfs_bmap_add_extent_unwritten_real(
159 xfs_inode_t *ip, /* incore inode pointer */
160 xfs_extnum_t idx, /* extent number to update/insert */
161 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
162 xfs_bmbt_irec_t *new, /* new data to add to file extents */
163 int *logflagsp, /* inode logging flags */
164 xfs_extdelta_t *delta); /* Change made to incore extents */
167 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
168 * It figures out where to ask the underlying allocator to put the new extent.
170 STATIC int /* error */
172 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
175 * Transform a btree format file with only one leaf node, where the
176 * extents list will fit in the inode, into an extents format file.
177 * Since the file extents are already in-core, all we have to do is
178 * give up the space for the btree root and pitch the leaf block.
180 STATIC int /* error */
181 xfs_bmap_btree_to_extents(
182 xfs_trans_t *tp, /* transaction pointer */
183 xfs_inode_t *ip, /* incore inode pointer */
184 xfs_btree_cur_t *cur, /* btree cursor */
185 int *logflagsp, /* inode logging flags */
186 int whichfork); /* data or attr fork */
190 * Check that the extents list for the inode ip is in the right order.
193 xfs_bmap_check_extents(
194 xfs_inode_t *ip, /* incore inode pointer */
195 int whichfork); /* data or attr fork */
199 * Called by xfs_bmapi to update file extent records and the btree
200 * after removing space (or undoing a delayed allocation).
202 STATIC int /* error */
204 xfs_inode_t *ip, /* incore inode pointer */
205 xfs_trans_t *tp, /* current trans pointer */
206 xfs_extnum_t idx, /* extent number to update/insert */
207 xfs_bmap_free_t *flist, /* list of extents to be freed */
208 xfs_btree_cur_t *cur, /* if null, not a btree */
209 xfs_bmbt_irec_t *new, /* new data to add to file extents */
210 int *logflagsp,/* inode logging flags */
211 xfs_extdelta_t *delta, /* Change made to incore extents */
212 int whichfork, /* data or attr fork */
213 int rsvd); /* OK to allocate reserved blocks */
216 * Remove the entry "free" from the free item list. Prev points to the
217 * previous entry, unless "free" is the head of the list.
221 xfs_bmap_free_t *flist, /* free item list header */
222 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
223 xfs_bmap_free_item_t *free); /* list item to be freed */
226 * Convert an extents-format file into a btree-format file.
227 * The new file will have a root block (in the inode) and a single child block.
229 STATIC int /* error */
230 xfs_bmap_extents_to_btree(
231 xfs_trans_t *tp, /* transaction pointer */
232 xfs_inode_t *ip, /* incore inode pointer */
233 xfs_fsblock_t *firstblock, /* first-block-allocated */
234 xfs_bmap_free_t *flist, /* blocks freed in xaction */
235 xfs_btree_cur_t **curp, /* cursor returned to caller */
236 int wasdel, /* converting a delayed alloc */
237 int *logflagsp, /* inode logging flags */
238 int whichfork); /* data or attr fork */
241 * Convert a local file to an extents file.
242 * This code is sort of bogus, since the file data needs to get
243 * logged so it won't be lost. The bmap-level manipulations are ok, though.
245 STATIC int /* error */
246 xfs_bmap_local_to_extents(
247 xfs_trans_t *tp, /* transaction pointer */
248 xfs_inode_t *ip, /* incore inode pointer */
249 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
250 xfs_extlen_t total, /* total blocks needed by transaction */
251 int *logflagsp, /* inode logging flags */
252 int whichfork); /* data or attr fork */
255 * Search the extents list for the inode, for the extent containing bno.
256 * If bno lies in a hole, point to the next entry. If bno lies past eof,
257 * *eofp will be set, and *prevp will contain the last entry (null if none).
258 * Else, *lastxp will be set to the index of the found
259 * entry; *gotp will contain the entry.
261 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
262 xfs_bmap_search_extents(
263 xfs_inode_t *ip, /* incore inode pointer */
264 xfs_fileoff_t bno, /* block number searched for */
265 int whichfork, /* data or attr fork */
266 int *eofp, /* out: end of file found */
267 xfs_extnum_t *lastxp, /* out: last extent index */
268 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
269 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
272 * Check the last inode extent to determine whether this allocation will result
273 * in blocks being allocated at the end of the file. When we allocate new data
274 * blocks at the end of the file which do not start at the previous data block,
275 * we will try to align the new blocks at stripe unit boundaries.
277 STATIC int /* error */
279 xfs_inode_t *ip, /* incore inode pointer */
280 xfs_fileoff_t off, /* file offset in fsblocks */
281 int whichfork, /* data or attribute fork */
282 char *aeof); /* return value */
284 #ifdef XFS_BMAP_TRACE
286 * Add a bmap trace buffer entry. Base routine for the others.
289 xfs_bmap_trace_addentry(
290 int opcode, /* operation */
291 char *fname, /* function name */
292 char *desc, /* operation description */
293 xfs_inode_t *ip, /* incore inode pointer */
294 xfs_extnum_t idx, /* index of entry(ies) */
295 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
296 xfs_bmbt_rec_t *r1, /* first record */
297 xfs_bmbt_rec_t *r2, /* second record or null */
298 int whichfork); /* data or attr fork */
301 * Add bmap trace entry prior to a call to xfs_iext_remove.
304 xfs_bmap_trace_delete(
305 char *fname, /* function name */
306 char *desc, /* operation description */
307 xfs_inode_t *ip, /* incore inode pointer */
308 xfs_extnum_t idx, /* index of entry(entries) deleted */
309 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
310 int whichfork); /* data or attr fork */
313 * Add bmap trace entry prior to a call to xfs_iext_insert, or
314 * reading in the extents list from the disk (in the btree).
317 xfs_bmap_trace_insert(
318 char *fname, /* function name */
319 char *desc, /* operation description */
320 xfs_inode_t *ip, /* incore inode pointer */
321 xfs_extnum_t idx, /* index of entry(entries) inserted */
322 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
323 xfs_bmbt_irec_t *r1, /* inserted record 1 */
324 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
325 int whichfork); /* data or attr fork */
328 * Add bmap trace entry after updating an extent record in place.
331 xfs_bmap_trace_post_update(
332 char *fname, /* function name */
333 char *desc, /* operation description */
334 xfs_inode_t *ip, /* incore inode pointer */
335 xfs_extnum_t idx, /* index of entry updated */
336 int whichfork); /* data or attr fork */
339 * Add bmap trace entry prior to updating an extent record in place.
342 xfs_bmap_trace_pre_update(
343 char *fname, /* function name */
344 char *desc, /* operation description */
345 xfs_inode_t *ip, /* incore inode pointer */
346 xfs_extnum_t idx, /* index of entry to be updated */
347 int whichfork); /* data or attr fork */
350 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
351 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
352 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
353 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
354 #endif /* XFS_BMAP_TRACE */
357 * Compute the worst-case number of indirect blocks that will be used
358 * for ip's delayed extent of length "len".
361 xfs_bmap_worst_indlen(
362 xfs_inode_t *ip, /* incore inode pointer */
363 xfs_filblks_t len); /* delayed extent length */
367 * Perform various validation checks on the values being returned
371 xfs_bmap_validate_ret(
375 xfs_bmbt_irec_t *mval,
379 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
382 #if defined(XFS_RW_TRACE)
391 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
392 #endif /* XFS_RW_TRACE */
399 xfs_fsblock_t blockno,
404 xfs_bmap_count_leaves(
411 xfs_bmap_disk_count_leaves(
415 xfs_bmbt_block_t *block,
420 * Bmap internal routines.
424 * Called from xfs_bmap_add_attrfork to handle btree format files.
426 STATIC int /* error */
427 xfs_bmap_add_attrfork_btree(
428 xfs_trans_t *tp, /* transaction pointer */
429 xfs_inode_t *ip, /* incore inode pointer */
430 xfs_fsblock_t *firstblock, /* first block allocated */
431 xfs_bmap_free_t *flist, /* blocks to free at commit */
432 int *flags) /* inode logging flags */
434 xfs_btree_cur_t *cur; /* btree cursor */
435 int error; /* error return value */
436 xfs_mount_t *mp; /* file system mount struct */
437 int stat; /* newroot status */
440 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
441 *flags |= XFS_ILOG_DBROOT;
443 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
445 cur->bc_private.b.flist = flist;
446 cur->bc_private.b.firstblock = *firstblock;
447 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
449 ASSERT(stat == 1); /* must be at least one entry */
450 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
453 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
454 return XFS_ERROR(ENOSPC);
456 *firstblock = cur->bc_private.b.firstblock;
457 cur->bc_private.b.allocated = 0;
458 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
462 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
467 * Called from xfs_bmap_add_attrfork to handle extents format files.
469 STATIC int /* error */
470 xfs_bmap_add_attrfork_extents(
471 xfs_trans_t *tp, /* transaction pointer */
472 xfs_inode_t *ip, /* incore inode pointer */
473 xfs_fsblock_t *firstblock, /* first block allocated */
474 xfs_bmap_free_t *flist, /* blocks to free at commit */
475 int *flags) /* inode logging flags */
477 xfs_btree_cur_t *cur; /* bmap btree cursor */
478 int error; /* error return value */
480 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
483 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
484 flags, XFS_DATA_FORK);
486 cur->bc_private.b.allocated = 0;
487 xfs_btree_del_cursor(cur,
488 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
494 * Called from xfs_bmap_add_attrfork to handle local format files.
496 STATIC int /* error */
497 xfs_bmap_add_attrfork_local(
498 xfs_trans_t *tp, /* transaction pointer */
499 xfs_inode_t *ip, /* incore inode pointer */
500 xfs_fsblock_t *firstblock, /* first block allocated */
501 xfs_bmap_free_t *flist, /* blocks to free at commit */
502 int *flags) /* inode logging flags */
504 xfs_da_args_t dargs; /* args for dir/attr code */
505 int error; /* error return value */
506 xfs_mount_t *mp; /* mount structure pointer */
508 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
510 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
512 memset(&dargs, 0, sizeof(dargs));
514 dargs.firstblock = firstblock;
516 dargs.total = mp->m_dirblkfsbs;
517 dargs.whichfork = XFS_DATA_FORK;
519 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
521 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
527 * Called by xfs_bmapi to update file extent records and the btree
528 * after allocating space (or doing a delayed allocation).
530 STATIC int /* error */
532 xfs_inode_t *ip, /* incore inode pointer */
533 xfs_extnum_t idx, /* extent number to update/insert */
534 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
535 xfs_bmbt_irec_t *new, /* new data to add to file extents */
536 xfs_fsblock_t *first, /* pointer to firstblock variable */
537 xfs_bmap_free_t *flist, /* list of extents to be freed */
538 int *logflagsp, /* inode logging flags */
539 xfs_extdelta_t *delta, /* Change made to incore extents */
540 int whichfork, /* data or attr fork */
541 int rsvd) /* OK to use reserved data blocks */
543 xfs_btree_cur_t *cur; /* btree cursor or null */
544 xfs_filblks_t da_new; /* new count del alloc blocks used */
545 xfs_filblks_t da_old; /* old count del alloc blocks used */
546 int error; /* error return value */
547 #ifdef XFS_BMAP_TRACE
548 static char fname[] = "xfs_bmap_add_extent";
550 xfs_ifork_t *ifp; /* inode fork ptr */
551 int logflags; /* returned value */
552 xfs_extnum_t nextents; /* number of extents in file now */
554 XFS_STATS_INC(xs_add_exlist);
556 ifp = XFS_IFORK_PTR(ip, whichfork);
557 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
558 ASSERT(idx <= nextents);
562 * This is the first extent added to a new/empty file.
563 * Special case this one, so other routines get to assume there are
564 * already extents in the list.
567 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
569 xfs_iext_insert(ifp, 0, 1, new);
572 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
573 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
574 logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
577 /* DELTA: single new extent */
579 if (delta->xed_startoff > new->br_startoff)
580 delta->xed_startoff = new->br_startoff;
581 if (delta->xed_blockcount <
582 new->br_startoff + new->br_blockcount)
583 delta->xed_blockcount = new->br_startoff +
588 * Any kind of new delayed allocation goes here.
590 else if (ISNULLSTARTBLOCK(new->br_startblock)) {
592 ASSERT((cur->bc_private.b.flags &
593 XFS_BTCUR_BPRV_WASDEL) == 0);
594 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
595 &logflags, delta, rsvd)))
599 * Real allocation off the end of the file.
601 else if (idx == nextents) {
603 ASSERT((cur->bc_private.b.flags &
604 XFS_BTCUR_BPRV_WASDEL) == 0);
605 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
606 &logflags, delta, whichfork)))
609 xfs_bmbt_irec_t prev; /* old extent at offset idx */
612 * Get the record referred to by idx.
614 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
616 * If it's a real allocation record, and the new allocation ends
617 * after the start of the referred to record, then we're filling
618 * in a delayed or unwritten allocation with a real one, or
619 * converting real back to unwritten.
621 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
622 new->br_startoff + new->br_blockcount > prev.br_startoff) {
623 if (prev.br_state != XFS_EXT_UNWRITTEN &&
624 ISNULLSTARTBLOCK(prev.br_startblock)) {
625 da_old = STARTBLOCKVAL(prev.br_startblock);
627 ASSERT(cur->bc_private.b.flags &
628 XFS_BTCUR_BPRV_WASDEL);
629 if ((error = xfs_bmap_add_extent_delay_real(ip,
630 idx, &cur, new, &da_new, first, flist,
631 &logflags, delta, rsvd)))
633 } else if (new->br_state == XFS_EXT_NORM) {
634 ASSERT(new->br_state == XFS_EXT_NORM);
635 if ((error = xfs_bmap_add_extent_unwritten_real(
636 ip, idx, &cur, new, &logflags, delta)))
639 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
640 if ((error = xfs_bmap_add_extent_unwritten_real(
641 ip, idx, &cur, new, &logflags, delta)))
644 ASSERT(*curp == cur || *curp == NULL);
647 * Otherwise we're filling in a hole with an allocation.
651 ASSERT((cur->bc_private.b.flags &
652 XFS_BTCUR_BPRV_WASDEL) == 0);
653 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
654 new, &logflags, delta, whichfork)))
659 ASSERT(*curp == cur || *curp == NULL);
661 * Convert to a btree if necessary.
663 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
664 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
665 int tmp_logflags; /* partial log flag return val */
668 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
669 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
670 logflags |= tmp_logflags;
675 * Adjust for changes in reserved delayed indirect blocks.
676 * Nothing to do for disk quotas here.
678 if (da_old || da_new) {
683 nblks += cur->bc_private.b.allocated;
684 ASSERT(nblks <= da_old);
686 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
687 (int)(da_old - nblks), rsvd);
690 * Clear out the allocated field, done with it now in any case.
693 cur->bc_private.b.allocated = 0;
699 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
701 *logflagsp = logflags;
706 * Called by xfs_bmap_add_extent to handle cases converting a delayed
707 * allocation to a real allocation.
709 STATIC int /* error */
710 xfs_bmap_add_extent_delay_real(
711 xfs_inode_t *ip, /* incore inode pointer */
712 xfs_extnum_t idx, /* extent number to update/insert */
713 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
714 xfs_bmbt_irec_t *new, /* new data to add to file extents */
715 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
716 xfs_fsblock_t *first, /* pointer to firstblock variable */
717 xfs_bmap_free_t *flist, /* list of extents to be freed */
718 int *logflagsp, /* inode logging flags */
719 xfs_extdelta_t *delta, /* Change made to incore extents */
720 int rsvd) /* OK to use reserved data block allocation */
722 xfs_btree_cur_t *cur; /* btree cursor */
723 int diff; /* temp value */
724 xfs_bmbt_rec_t *ep; /* extent entry for idx */
725 int error; /* error return value */
726 #ifdef XFS_BMAP_TRACE
727 static char fname[] = "xfs_bmap_add_extent_delay_real";
729 int i; /* temp state */
730 xfs_ifork_t *ifp; /* inode fork pointer */
731 xfs_fileoff_t new_endoff; /* end offset of new entry */
732 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
733 /* left is 0, right is 1, prev is 2 */
734 int rval=0; /* return value (logging flags) */
735 int state = 0;/* state bits, accessed thru macros */
736 xfs_filblks_t temp=0; /* value for dnew calculations */
737 xfs_filblks_t temp2=0;/* value for dnew calculations */
738 int tmp_rval; /* partial logging flags */
739 enum { /* bit number definitions for state */
740 LEFT_CONTIG, RIGHT_CONTIG,
741 LEFT_FILLING, RIGHT_FILLING,
742 LEFT_DELAY, RIGHT_DELAY,
743 LEFT_VALID, RIGHT_VALID
749 #define MASK(b) (1 << (b))
750 #define MASK2(a,b) (MASK(a) | MASK(b))
751 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
752 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
753 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
754 #define STATE_TEST(b) (state & MASK(b))
755 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
756 ((state &= ~MASK(b)), 0))
757 #define SWITCH_STATE \
758 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
761 * Set up a bunch of variables to make the tests simpler.
764 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
765 ep = xfs_iext_get_ext(ifp, idx);
766 xfs_bmbt_get_all(ep, &PREV);
767 new_endoff = new->br_startoff + new->br_blockcount;
768 ASSERT(PREV.br_startoff <= new->br_startoff);
769 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
771 * Set flags determining what part of the previous delayed allocation
772 * extent is being replaced by a real allocation.
774 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
775 STATE_SET(RIGHT_FILLING,
776 PREV.br_startoff + PREV.br_blockcount == new_endoff);
778 * Check and set flags if this segment has a left neighbor.
779 * Don't set contiguous if the combined extent would be too large.
781 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
782 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
783 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
785 STATE_SET(LEFT_CONTIG,
786 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
787 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
788 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
789 LEFT.br_state == new->br_state &&
790 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
792 * Check and set flags if this segment has a right neighbor.
793 * Don't set contiguous if the combined extent would be too large.
794 * Also check for all-three-contiguous being too large.
796 if (STATE_SET_TEST(RIGHT_VALID,
798 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
799 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
800 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
802 STATE_SET(RIGHT_CONTIG,
803 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
804 new_endoff == RIGHT.br_startoff &&
805 new->br_startblock + new->br_blockcount ==
806 RIGHT.br_startblock &&
807 new->br_state == RIGHT.br_state &&
808 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
809 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
810 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
811 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
815 * Switch out based on the FILLING and CONTIG state bits.
817 switch (SWITCH_STATE) {
819 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
821 * Filling in all of a previously delayed allocation extent.
822 * The left and right neighbors are both contiguous with new.
824 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
826 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
827 LEFT.br_blockcount + PREV.br_blockcount +
828 RIGHT.br_blockcount);
829 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
831 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
833 xfs_iext_remove(ifp, idx, 2);
834 ip->i_df.if_lastex = idx - 1;
835 ip->i_d.di_nextents--;
837 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
839 rval = XFS_ILOG_CORE;
840 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
842 RIGHT.br_blockcount, &i)))
845 if ((error = xfs_bmbt_delete(cur, &i)))
848 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
851 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
855 RIGHT.br_blockcount, LEFT.br_state)))
859 /* DELTA: Three in-core extents are replaced by one. */
860 temp = LEFT.br_startoff;
861 temp2 = LEFT.br_blockcount +
866 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
868 * Filling in all of a previously delayed allocation extent.
869 * The left neighbor is contiguous, the right is not.
871 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
873 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
874 LEFT.br_blockcount + PREV.br_blockcount);
875 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
877 ip->i_df.if_lastex = idx - 1;
878 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
880 xfs_iext_remove(ifp, idx, 1);
882 rval = XFS_ILOG_DEXT;
885 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
886 LEFT.br_startblock, LEFT.br_blockcount,
890 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
893 PREV.br_blockcount, LEFT.br_state)))
897 /* DELTA: Two in-core extents are replaced by one. */
898 temp = LEFT.br_startoff;
899 temp2 = LEFT.br_blockcount +
903 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
905 * Filling in all of a previously delayed allocation extent.
906 * The right neighbor is contiguous, the left is not.
908 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
910 xfs_bmbt_set_startblock(ep, new->br_startblock);
911 xfs_bmbt_set_blockcount(ep,
912 PREV.br_blockcount + RIGHT.br_blockcount);
913 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
915 ip->i_df.if_lastex = idx;
916 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
918 xfs_iext_remove(ifp, idx + 1, 1);
920 rval = XFS_ILOG_DEXT;
923 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
925 RIGHT.br_blockcount, &i)))
928 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
931 RIGHT.br_blockcount, PREV.br_state)))
935 /* DELTA: Two in-core extents are replaced by one. */
936 temp = PREV.br_startoff;
937 temp2 = PREV.br_blockcount +
941 case MASK2(LEFT_FILLING, RIGHT_FILLING):
943 * Filling in all of a previously delayed allocation extent.
944 * Neither the left nor right neighbors are contiguous with
947 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
949 xfs_bmbt_set_startblock(ep, new->br_startblock);
950 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
952 ip->i_df.if_lastex = idx;
953 ip->i_d.di_nextents++;
955 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
957 rval = XFS_ILOG_CORE;
958 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
959 new->br_startblock, new->br_blockcount,
963 cur->bc_rec.b.br_state = XFS_EXT_NORM;
964 if ((error = xfs_bmbt_insert(cur, &i)))
969 /* DELTA: The in-core extent described by new changed type. */
970 temp = new->br_startoff;
971 temp2 = new->br_blockcount;
974 case MASK2(LEFT_FILLING, LEFT_CONTIG):
976 * Filling in the first part of a previous delayed allocation.
977 * The left neighbor is contiguous.
979 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
981 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
982 LEFT.br_blockcount + new->br_blockcount);
983 xfs_bmbt_set_startoff(ep,
984 PREV.br_startoff + new->br_blockcount);
985 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
987 temp = PREV.br_blockcount - new->br_blockcount;
988 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
990 xfs_bmbt_set_blockcount(ep, temp);
991 ip->i_df.if_lastex = idx - 1;
993 rval = XFS_ILOG_DEXT;
996 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
997 LEFT.br_startblock, LEFT.br_blockcount,
1001 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1003 LEFT.br_blockcount +
1008 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1009 STARTBLOCKVAL(PREV.br_startblock));
1010 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1011 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1014 /* DELTA: The boundary between two in-core extents moved. */
1015 temp = LEFT.br_startoff;
1016 temp2 = LEFT.br_blockcount +
1020 case MASK(LEFT_FILLING):
1022 * Filling in the first part of a previous delayed allocation.
1023 * The left neighbor is not contiguous.
1025 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1026 xfs_bmbt_set_startoff(ep, new_endoff);
1027 temp = PREV.br_blockcount - new->br_blockcount;
1028 xfs_bmbt_set_blockcount(ep, temp);
1029 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1031 xfs_iext_insert(ifp, idx, 1, new);
1032 ip->i_df.if_lastex = idx;
1033 ip->i_d.di_nextents++;
1035 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1037 rval = XFS_ILOG_CORE;
1038 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1039 new->br_startblock, new->br_blockcount,
1043 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1044 if ((error = xfs_bmbt_insert(cur, &i)))
1048 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1049 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1050 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1051 first, flist, &cur, 1, &tmp_rval,
1057 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1058 STARTBLOCKVAL(PREV.br_startblock) -
1059 (cur ? cur->bc_private.b.allocated : 0));
1060 ep = xfs_iext_get_ext(ifp, idx + 1);
1061 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1062 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1065 /* DELTA: One in-core extent is split in two. */
1066 temp = PREV.br_startoff;
1067 temp2 = PREV.br_blockcount;
1070 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1072 * Filling in the last part of a previous delayed allocation.
1073 * The right neighbor is contiguous with the new allocation.
1075 temp = PREV.br_blockcount - new->br_blockcount;
1076 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1078 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1080 xfs_bmbt_set_blockcount(ep, temp);
1081 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1082 new->br_startoff, new->br_startblock,
1083 new->br_blockcount + RIGHT.br_blockcount,
1085 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1087 ip->i_df.if_lastex = idx + 1;
1089 rval = XFS_ILOG_DEXT;
1092 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1093 RIGHT.br_startblock,
1094 RIGHT.br_blockcount, &i)))
1097 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1099 new->br_blockcount +
1100 RIGHT.br_blockcount,
1104 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1105 STARTBLOCKVAL(PREV.br_startblock));
1106 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1107 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1110 /* DELTA: The boundary between two in-core extents moved. */
1111 temp = PREV.br_startoff;
1112 temp2 = PREV.br_blockcount +
1113 RIGHT.br_blockcount;
1116 case MASK(RIGHT_FILLING):
1118 * Filling in the last part of a previous delayed allocation.
1119 * The right neighbor is not contiguous.
1121 temp = PREV.br_blockcount - new->br_blockcount;
1122 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1123 xfs_bmbt_set_blockcount(ep, temp);
1124 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1125 new, NULL, XFS_DATA_FORK);
1126 xfs_iext_insert(ifp, idx + 1, 1, new);
1127 ip->i_df.if_lastex = idx + 1;
1128 ip->i_d.di_nextents++;
1130 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1132 rval = XFS_ILOG_CORE;
1133 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1134 new->br_startblock, new->br_blockcount,
1138 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1139 if ((error = xfs_bmbt_insert(cur, &i)))
1143 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1144 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1145 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1146 first, flist, &cur, 1, &tmp_rval,
1152 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1153 STARTBLOCKVAL(PREV.br_startblock) -
1154 (cur ? cur->bc_private.b.allocated : 0));
1155 ep = xfs_iext_get_ext(ifp, idx);
1156 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1157 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1159 /* DELTA: One in-core extent is split in two. */
1160 temp = PREV.br_startoff;
1161 temp2 = PREV.br_blockcount;
1166 * Filling in the middle part of a previous delayed allocation.
1167 * Contiguity is impossible here.
1168 * This case is avoided almost all the time.
1170 temp = new->br_startoff - PREV.br_startoff;
1171 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1172 xfs_bmbt_set_blockcount(ep, temp);
1174 r[1].br_startoff = new_endoff;
1175 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1176 r[1].br_blockcount = temp2;
1177 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1179 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1180 ip->i_df.if_lastex = idx + 1;
1181 ip->i_d.di_nextents++;
1183 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1185 rval = XFS_ILOG_CORE;
1186 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1187 new->br_startblock, new->br_blockcount,
1191 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1192 if ((error = xfs_bmbt_insert(cur, &i)))
1196 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1197 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1198 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1199 first, flist, &cur, 1, &tmp_rval,
1205 temp = xfs_bmap_worst_indlen(ip, temp);
1206 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1207 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1208 (cur ? cur->bc_private.b.allocated : 0));
1210 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1212 * Ick gross gag me with a spoon.
1214 ASSERT(0); /* want to see if this ever happens! */
1220 !xfs_mod_incore_sb(ip->i_mount,
1221 XFS_SBS_FDBLOCKS, -diff, rsvd))
1228 !xfs_mod_incore_sb(ip->i_mount,
1229 XFS_SBS_FDBLOCKS, -diff, rsvd))
1234 ep = xfs_iext_get_ext(ifp, idx);
1235 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1236 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1237 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1239 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1240 NULLSTARTBLOCK((int)temp2));
1241 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1243 *dnew = temp + temp2;
1244 /* DELTA: One in-core extent is split in three. */
1245 temp = PREV.br_startoff;
1246 temp2 = PREV.br_blockcount;
1249 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1250 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1251 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1252 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1253 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1254 case MASK(LEFT_CONTIG):
1255 case MASK(RIGHT_CONTIG):
1257 * These cases are all impossible.
1264 if (delta->xed_startoff > temp)
1265 delta->xed_startoff = temp;
1266 if (delta->xed_blockcount < temp2)
1267 delta->xed_blockcount = temp2;
1281 #undef STATE_SET_TEST
1286 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1287 * allocation to a real allocation or vice versa.
1289 STATIC int /* error */
1290 xfs_bmap_add_extent_unwritten_real(
1291 xfs_inode_t *ip, /* incore inode pointer */
1292 xfs_extnum_t idx, /* extent number to update/insert */
1293 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1294 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1295 int *logflagsp, /* inode logging flags */
1296 xfs_extdelta_t *delta) /* Change made to incore extents */
1298 xfs_btree_cur_t *cur; /* btree cursor */
1299 xfs_bmbt_rec_t *ep; /* extent entry for idx */
1300 int error; /* error return value */
1301 #ifdef XFS_BMAP_TRACE
1302 static char fname[] = "xfs_bmap_add_extent_unwritten_real";
1304 int i; /* temp state */
1305 xfs_ifork_t *ifp; /* inode fork pointer */
1306 xfs_fileoff_t new_endoff; /* end offset of new entry */
1307 xfs_exntst_t newext; /* new extent state */
1308 xfs_exntst_t oldext; /* old extent state */
1309 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1310 /* left is 0, right is 1, prev is 2 */
1311 int rval=0; /* return value (logging flags) */
1312 int state = 0;/* state bits, accessed thru macros */
1313 xfs_filblks_t temp=0;
1314 xfs_filblks_t temp2=0;
1315 enum { /* bit number definitions for state */
1316 LEFT_CONTIG, RIGHT_CONTIG,
1317 LEFT_FILLING, RIGHT_FILLING,
1318 LEFT_DELAY, RIGHT_DELAY,
1319 LEFT_VALID, RIGHT_VALID
1325 #define MASK(b) (1 << (b))
1326 #define MASK2(a,b) (MASK(a) | MASK(b))
1327 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
1328 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
1329 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1330 #define STATE_TEST(b) (state & MASK(b))
1331 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1332 ((state &= ~MASK(b)), 0))
1333 #define SWITCH_STATE \
1334 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1337 * Set up a bunch of variables to make the tests simpler.
1341 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1342 ep = xfs_iext_get_ext(ifp, idx);
1343 xfs_bmbt_get_all(ep, &PREV);
1344 newext = new->br_state;
1345 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1346 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1347 ASSERT(PREV.br_state == oldext);
1348 new_endoff = new->br_startoff + new->br_blockcount;
1349 ASSERT(PREV.br_startoff <= new->br_startoff);
1350 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1352 * Set flags determining what part of the previous oldext allocation
1353 * extent is being replaced by a newext allocation.
1355 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1356 STATE_SET(RIGHT_FILLING,
1357 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1359 * Check and set flags if this segment has a left neighbor.
1360 * Don't set contiguous if the combined extent would be too large.
1362 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1363 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1364 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1366 STATE_SET(LEFT_CONTIG,
1367 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1368 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1369 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1370 LEFT.br_state == newext &&
1371 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1373 * Check and set flags if this segment has a right neighbor.
1374 * Don't set contiguous if the combined extent would be too large.
1375 * Also check for all-three-contiguous being too large.
1377 if (STATE_SET_TEST(RIGHT_VALID,
1379 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1380 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1381 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1383 STATE_SET(RIGHT_CONTIG,
1384 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1385 new_endoff == RIGHT.br_startoff &&
1386 new->br_startblock + new->br_blockcount ==
1387 RIGHT.br_startblock &&
1388 newext == RIGHT.br_state &&
1389 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1390 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1391 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1392 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1395 * Switch out based on the FILLING and CONTIG state bits.
1397 switch (SWITCH_STATE) {
1399 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1401 * Setting all of a previous oldext extent to newext.
1402 * The left and right neighbors are both contiguous with new.
1404 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1406 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1407 LEFT.br_blockcount + PREV.br_blockcount +
1408 RIGHT.br_blockcount);
1409 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1411 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1413 xfs_iext_remove(ifp, idx, 2);
1414 ip->i_df.if_lastex = idx - 1;
1415 ip->i_d.di_nextents -= 2;
1417 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1419 rval = XFS_ILOG_CORE;
1420 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1421 RIGHT.br_startblock,
1422 RIGHT.br_blockcount, &i)))
1425 if ((error = xfs_bmbt_delete(cur, &i)))
1428 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1431 if ((error = xfs_bmbt_delete(cur, &i)))
1434 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1437 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1439 LEFT.br_blockcount + PREV.br_blockcount +
1440 RIGHT.br_blockcount, LEFT.br_state)))
1443 /* DELTA: Three in-core extents are replaced by one. */
1444 temp = LEFT.br_startoff;
1445 temp2 = LEFT.br_blockcount +
1446 PREV.br_blockcount +
1447 RIGHT.br_blockcount;
1450 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1452 * Setting all of a previous oldext extent to newext.
1453 * The left neighbor is contiguous, the right is not.
1455 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1457 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1458 LEFT.br_blockcount + PREV.br_blockcount);
1459 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1461 ip->i_df.if_lastex = idx - 1;
1462 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1464 xfs_iext_remove(ifp, idx, 1);
1465 ip->i_d.di_nextents--;
1467 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1469 rval = XFS_ILOG_CORE;
1470 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1471 PREV.br_startblock, PREV.br_blockcount,
1475 if ((error = xfs_bmbt_delete(cur, &i)))
1478 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1481 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1483 LEFT.br_blockcount + PREV.br_blockcount,
1487 /* DELTA: Two in-core extents are replaced by one. */
1488 temp = LEFT.br_startoff;
1489 temp2 = LEFT.br_blockcount +
1493 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1495 * Setting all of a previous oldext extent to newext.
1496 * The right neighbor is contiguous, the left is not.
1498 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1500 xfs_bmbt_set_blockcount(ep,
1501 PREV.br_blockcount + RIGHT.br_blockcount);
1502 xfs_bmbt_set_state(ep, newext);
1503 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1505 ip->i_df.if_lastex = idx;
1506 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1508 xfs_iext_remove(ifp, idx + 1, 1);
1509 ip->i_d.di_nextents--;
1511 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1513 rval = XFS_ILOG_CORE;
1514 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1515 RIGHT.br_startblock,
1516 RIGHT.br_blockcount, &i)))
1519 if ((error = xfs_bmbt_delete(cur, &i)))
1522 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1525 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1527 new->br_blockcount + RIGHT.br_blockcount,
1531 /* DELTA: Two in-core extents are replaced by one. */
1532 temp = PREV.br_startoff;
1533 temp2 = PREV.br_blockcount +
1534 RIGHT.br_blockcount;
1537 case MASK2(LEFT_FILLING, RIGHT_FILLING):
1539 * Setting all of a previous oldext extent to newext.
1540 * Neither the left nor right neighbors are contiguous with
1543 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1545 xfs_bmbt_set_state(ep, newext);
1546 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1548 ip->i_df.if_lastex = idx;
1550 rval = XFS_ILOG_DEXT;
1553 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1554 new->br_startblock, new->br_blockcount,
1558 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1559 new->br_startblock, new->br_blockcount,
1563 /* DELTA: The in-core extent described by new changed type. */
1564 temp = new->br_startoff;
1565 temp2 = new->br_blockcount;
1568 case MASK2(LEFT_FILLING, LEFT_CONTIG):
1570 * Setting the first part of a previous oldext extent to newext.
1571 * The left neighbor is contiguous.
1573 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1575 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1576 LEFT.br_blockcount + new->br_blockcount);
1577 xfs_bmbt_set_startoff(ep,
1578 PREV.br_startoff + new->br_blockcount);
1579 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1581 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1583 xfs_bmbt_set_startblock(ep,
1584 new->br_startblock + new->br_blockcount);
1585 xfs_bmbt_set_blockcount(ep,
1586 PREV.br_blockcount - new->br_blockcount);
1587 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1589 ip->i_df.if_lastex = idx - 1;
1591 rval = XFS_ILOG_DEXT;
1594 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1595 PREV.br_startblock, PREV.br_blockcount,
1599 if ((error = xfs_bmbt_update(cur,
1600 PREV.br_startoff + new->br_blockcount,
1601 PREV.br_startblock + new->br_blockcount,
1602 PREV.br_blockcount - new->br_blockcount,
1605 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1607 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1609 LEFT.br_blockcount + new->br_blockcount,
1613 /* DELTA: The boundary between two in-core extents moved. */
1614 temp = LEFT.br_startoff;
1615 temp2 = LEFT.br_blockcount +
1619 case MASK(LEFT_FILLING):
1621 * Setting the first part of a previous oldext extent to newext.
1622 * The left neighbor is not contiguous.
1624 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1625 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1626 xfs_bmbt_set_startoff(ep, new_endoff);
1627 xfs_bmbt_set_blockcount(ep,
1628 PREV.br_blockcount - new->br_blockcount);
1629 xfs_bmbt_set_startblock(ep,
1630 new->br_startblock + new->br_blockcount);
1631 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1632 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1634 xfs_iext_insert(ifp, idx, 1, new);
1635 ip->i_df.if_lastex = idx;
1636 ip->i_d.di_nextents++;
1638 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1640 rval = XFS_ILOG_CORE;
1641 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1642 PREV.br_startblock, PREV.br_blockcount,
1646 if ((error = xfs_bmbt_update(cur,
1647 PREV.br_startoff + new->br_blockcount,
1648 PREV.br_startblock + new->br_blockcount,
1649 PREV.br_blockcount - new->br_blockcount,
1652 cur->bc_rec.b = *new;
1653 if ((error = xfs_bmbt_insert(cur, &i)))
1657 /* DELTA: One in-core extent is split in two. */
1658 temp = PREV.br_startoff;
1659 temp2 = PREV.br_blockcount;
1662 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1664 * Setting the last part of a previous oldext extent to newext.
1665 * The right neighbor is contiguous with the new allocation.
1667 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1669 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1671 xfs_bmbt_set_blockcount(ep,
1672 PREV.br_blockcount - new->br_blockcount);
1673 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1675 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1676 new->br_startoff, new->br_startblock,
1677 new->br_blockcount + RIGHT.br_blockcount, newext);
1678 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1680 ip->i_df.if_lastex = idx + 1;
1682 rval = XFS_ILOG_DEXT;
1685 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1687 PREV.br_blockcount, &i)))
1690 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1692 PREV.br_blockcount - new->br_blockcount,
1695 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1697 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1699 new->br_blockcount + RIGHT.br_blockcount,
1703 /* DELTA: The boundary between two in-core extents moved. */
1704 temp = PREV.br_startoff;
1705 temp2 = PREV.br_blockcount +
1706 RIGHT.br_blockcount;
1709 case MASK(RIGHT_FILLING):
1711 * Setting the last part of a previous oldext extent to newext.
1712 * The right neighbor is not contiguous.
1714 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1715 xfs_bmbt_set_blockcount(ep,
1716 PREV.br_blockcount - new->br_blockcount);
1717 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1718 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1719 new, NULL, XFS_DATA_FORK);
1720 xfs_iext_insert(ifp, idx + 1, 1, new);
1721 ip->i_df.if_lastex = idx + 1;
1722 ip->i_d.di_nextents++;
1724 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1726 rval = XFS_ILOG_CORE;
1727 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1728 PREV.br_startblock, PREV.br_blockcount,
1732 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1734 PREV.br_blockcount - new->br_blockcount,
1737 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1738 new->br_startblock, new->br_blockcount,
1742 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1743 if ((error = xfs_bmbt_insert(cur, &i)))
1747 /* DELTA: One in-core extent is split in two. */
1748 temp = PREV.br_startoff;
1749 temp2 = PREV.br_blockcount;
1754 * Setting the middle part of a previous oldext extent to
1755 * newext. Contiguity is impossible here.
1756 * One extent becomes three extents.
1758 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1759 xfs_bmbt_set_blockcount(ep,
1760 new->br_startoff - PREV.br_startoff);
1761 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1763 r[1].br_startoff = new_endoff;
1764 r[1].br_blockcount =
1765 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1766 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1767 r[1].br_state = oldext;
1768 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1770 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1771 ip->i_df.if_lastex = idx + 1;
1772 ip->i_d.di_nextents += 2;
1774 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1776 rval = XFS_ILOG_CORE;
1777 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1778 PREV.br_startblock, PREV.br_blockcount,
1782 /* new right extent - oldext */
1783 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1784 r[1].br_startblock, r[1].br_blockcount,
1787 /* new left extent - oldext */
1788 PREV.br_blockcount =
1789 new->br_startoff - PREV.br_startoff;
1790 cur->bc_rec.b = PREV;
1791 if ((error = xfs_bmbt_insert(cur, &i)))
1794 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1797 /* new middle extent - newext */
1798 cur->bc_rec.b = *new;
1799 if ((error = xfs_bmbt_insert(cur, &i)))
1803 /* DELTA: One in-core extent is split in three. */
1804 temp = PREV.br_startoff;
1805 temp2 = PREV.br_blockcount;
1808 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1809 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1810 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1811 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1812 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1813 case MASK(LEFT_CONTIG):
1814 case MASK(RIGHT_CONTIG):
1816 * These cases are all impossible.
1823 if (delta->xed_startoff > temp)
1824 delta->xed_startoff = temp;
1825 if (delta->xed_blockcount < temp2)
1826 delta->xed_blockcount = temp2;
1840 #undef STATE_SET_TEST
1845 * Called by xfs_bmap_add_extent to handle cases converting a hole
1846 * to a delayed allocation.
1849 STATIC int /* error */
1850 xfs_bmap_add_extent_hole_delay(
1851 xfs_inode_t *ip, /* incore inode pointer */
1852 xfs_extnum_t idx, /* extent number to update/insert */
1853 xfs_btree_cur_t *cur, /* if null, not a btree */
1854 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1855 int *logflagsp, /* inode logging flags */
1856 xfs_extdelta_t *delta, /* Change made to incore extents */
1857 int rsvd) /* OK to allocate reserved blocks */
1859 xfs_bmbt_rec_t *ep; /* extent record for idx */
1860 #ifdef XFS_BMAP_TRACE
1861 static char fname[] = "xfs_bmap_add_extent_hole_delay";
1863 xfs_ifork_t *ifp; /* inode fork pointer */
1864 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1865 xfs_filblks_t newlen=0; /* new indirect size */
1866 xfs_filblks_t oldlen=0; /* old indirect size */
1867 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1868 int state; /* state bits, accessed thru macros */
1869 xfs_filblks_t temp=0; /* temp for indirect calculations */
1870 xfs_filblks_t temp2=0;
1871 enum { /* bit number definitions for state */
1872 LEFT_CONTIG, RIGHT_CONTIG,
1873 LEFT_DELAY, RIGHT_DELAY,
1874 LEFT_VALID, RIGHT_VALID
1877 #define MASK(b) (1 << (b))
1878 #define MASK2(a,b) (MASK(a) | MASK(b))
1879 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1880 #define STATE_TEST(b) (state & MASK(b))
1881 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1882 ((state &= ~MASK(b)), 0))
1883 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1885 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1886 ep = xfs_iext_get_ext(ifp, idx);
1888 ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1890 * Check and set flags if this segment has a left neighbor
1892 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1893 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1894 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1897 * Check and set flags if the current (right) segment exists.
1898 * If it doesn't exist, we're converting the hole at end-of-file.
1900 if (STATE_SET_TEST(RIGHT_VALID,
1902 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1903 xfs_bmbt_get_all(ep, &right);
1904 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1907 * Set contiguity flags on the left and right neighbors.
1908 * Don't let extents get too large, even if the pieces are contiguous.
1910 STATE_SET(LEFT_CONTIG,
1911 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1912 left.br_startoff + left.br_blockcount == new->br_startoff &&
1913 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1914 STATE_SET(RIGHT_CONTIG,
1915 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1916 new->br_startoff + new->br_blockcount == right.br_startoff &&
1917 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1918 (!STATE_TEST(LEFT_CONTIG) ||
1919 (left.br_blockcount + new->br_blockcount +
1920 right.br_blockcount <= MAXEXTLEN)));
1922 * Switch out based on the contiguity flags.
1924 switch (SWITCH_STATE) {
1926 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1928 * New allocation is contiguous with delayed allocations
1929 * on the left and on the right.
1930 * Merge all three into a single extent record.
1932 temp = left.br_blockcount + new->br_blockcount +
1933 right.br_blockcount;
1934 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1936 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1937 oldlen = STARTBLOCKVAL(left.br_startblock) +
1938 STARTBLOCKVAL(new->br_startblock) +
1939 STARTBLOCKVAL(right.br_startblock);
1940 newlen = xfs_bmap_worst_indlen(ip, temp);
1941 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1942 NULLSTARTBLOCK((int)newlen));
1943 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1945 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1947 xfs_iext_remove(ifp, idx, 1);
1948 ip->i_df.if_lastex = idx - 1;
1949 /* DELTA: Two in-core extents were replaced by one. */
1951 temp = left.br_startoff;
1954 case MASK(LEFT_CONTIG):
1956 * New allocation is contiguous with a delayed allocation
1958 * Merge the new allocation with the left neighbor.
1960 temp = left.br_blockcount + new->br_blockcount;
1961 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1963 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1964 oldlen = STARTBLOCKVAL(left.br_startblock) +
1965 STARTBLOCKVAL(new->br_startblock);
1966 newlen = xfs_bmap_worst_indlen(ip, temp);
1967 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1968 NULLSTARTBLOCK((int)newlen));
1969 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1971 ip->i_df.if_lastex = idx - 1;
1972 /* DELTA: One in-core extent grew into a hole. */
1974 temp = left.br_startoff;
1977 case MASK(RIGHT_CONTIG):
1979 * New allocation is contiguous with a delayed allocation
1981 * Merge the new allocation with the right neighbor.
1983 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1984 temp = new->br_blockcount + right.br_blockcount;
1985 oldlen = STARTBLOCKVAL(new->br_startblock) +
1986 STARTBLOCKVAL(right.br_startblock);
1987 newlen = xfs_bmap_worst_indlen(ip, temp);
1988 xfs_bmbt_set_allf(ep, new->br_startoff,
1989 NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1990 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1991 ip->i_df.if_lastex = idx;
1992 /* DELTA: One in-core extent grew into a hole. */
1994 temp = new->br_startoff;
1999 * New allocation is not contiguous with another
2000 * delayed allocation.
2001 * Insert a new entry.
2003 oldlen = newlen = 0;
2004 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2006 xfs_iext_insert(ifp, idx, 1, new);
2007 ip->i_df.if_lastex = idx;
2008 /* DELTA: A new in-core extent was added in a hole. */
2009 temp2 = new->br_blockcount;
2010 temp = new->br_startoff;
2013 if (oldlen != newlen) {
2014 ASSERT(oldlen > newlen);
2015 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
2016 (int)(oldlen - newlen), rsvd);
2018 * Nothing to do for disk quota accounting here.
2023 if (delta->xed_startoff > temp)
2024 delta->xed_startoff = temp;
2025 if (delta->xed_blockcount < temp2)
2026 delta->xed_blockcount = temp2;
2034 #undef STATE_SET_TEST
2039 * Called by xfs_bmap_add_extent to handle cases converting a hole
2040 * to a real allocation.
2042 STATIC int /* error */
2043 xfs_bmap_add_extent_hole_real(
2044 xfs_inode_t *ip, /* incore inode pointer */
2045 xfs_extnum_t idx, /* extent number to update/insert */
2046 xfs_btree_cur_t *cur, /* if null, not a btree */
2047 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2048 int *logflagsp, /* inode logging flags */
2049 xfs_extdelta_t *delta, /* Change made to incore extents */
2050 int whichfork) /* data or attr fork */
2052 xfs_bmbt_rec_t *ep; /* pointer to extent entry ins. point */
2053 int error; /* error return value */
2054 #ifdef XFS_BMAP_TRACE
2055 static char fname[] = "xfs_bmap_add_extent_hole_real";
2057 int i; /* temp state */
2058 xfs_ifork_t *ifp; /* inode fork pointer */
2059 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2060 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2061 int rval=0; /* return value (logging flags) */
2062 int state; /* state bits, accessed thru macros */
2063 xfs_filblks_t temp=0;
2064 xfs_filblks_t temp2=0;
2065 enum { /* bit number definitions for state */
2066 LEFT_CONTIG, RIGHT_CONTIG,
2067 LEFT_DELAY, RIGHT_DELAY,
2068 LEFT_VALID, RIGHT_VALID
2071 #define MASK(b) (1 << (b))
2072 #define MASK2(a,b) (MASK(a) | MASK(b))
2073 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2074 #define STATE_TEST(b) (state & MASK(b))
2075 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
2076 ((state &= ~MASK(b)), 0))
2077 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2079 ifp = XFS_IFORK_PTR(ip, whichfork);
2080 ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2081 ep = xfs_iext_get_ext(ifp, idx);
2084 * Check and set flags if this segment has a left neighbor.
2086 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2087 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2088 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2091 * Check and set flags if this segment has a current value.
2092 * Not true if we're inserting into the "hole" at eof.
2094 if (STATE_SET_TEST(RIGHT_VALID,
2096 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2097 xfs_bmbt_get_all(ep, &right);
2098 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2101 * We're inserting a real allocation between "left" and "right".
2102 * Set the contiguity flags. Don't let extents get too large.
2104 STATE_SET(LEFT_CONTIG,
2105 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2106 left.br_startoff + left.br_blockcount == new->br_startoff &&
2107 left.br_startblock + left.br_blockcount == new->br_startblock &&
2108 left.br_state == new->br_state &&
2109 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2110 STATE_SET(RIGHT_CONTIG,
2111 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2112 new->br_startoff + new->br_blockcount == right.br_startoff &&
2113 new->br_startblock + new->br_blockcount ==
2114 right.br_startblock &&
2115 new->br_state == right.br_state &&
2116 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2117 (!STATE_TEST(LEFT_CONTIG) ||
2118 left.br_blockcount + new->br_blockcount +
2119 right.br_blockcount <= MAXEXTLEN));
2123 * Select which case we're in here, and implement it.
2125 switch (SWITCH_STATE) {
2127 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2129 * New allocation is contiguous with real allocations on the
2130 * left and on the right.
2131 * Merge all three into a single extent record.
2133 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2135 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2136 left.br_blockcount + new->br_blockcount +
2137 right.br_blockcount);
2138 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2140 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2142 xfs_iext_remove(ifp, idx, 1);
2143 ifp->if_lastex = idx - 1;
2144 XFS_IFORK_NEXT_SET(ip, whichfork,
2145 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2147 rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2149 rval = XFS_ILOG_CORE;
2150 if ((error = xfs_bmbt_lookup_eq(cur,
2152 right.br_startblock,
2153 right.br_blockcount, &i)))
2156 if ((error = xfs_bmbt_delete(cur, &i)))
2159 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2162 if ((error = xfs_bmbt_update(cur, left.br_startoff,
2164 left.br_blockcount +
2165 new->br_blockcount +
2166 right.br_blockcount,
2170 /* DELTA: Two in-core extents were replaced by one. */
2171 temp = left.br_startoff;
2172 temp2 = left.br_blockcount +
2173 new->br_blockcount +
2174 right.br_blockcount;
2177 case MASK(LEFT_CONTIG):
2179 * New allocation is contiguous with a real allocation
2181 * Merge the new allocation with the left neighbor.
2183 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2184 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2185 left.br_blockcount + new->br_blockcount);
2186 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2187 ifp->if_lastex = idx - 1;
2189 rval = XFS_ILOG_FEXT(whichfork);
2192 if ((error = xfs_bmbt_lookup_eq(cur,
2195 left.br_blockcount, &i)))
2198 if ((error = xfs_bmbt_update(cur, left.br_startoff,
2200 left.br_blockcount +
2205 /* DELTA: One in-core extent grew. */
2206 temp = left.br_startoff;
2207 temp2 = left.br_blockcount +
2211 case MASK(RIGHT_CONTIG):
2213 * New allocation is contiguous with a real allocation
2215 * Merge the new allocation with the right neighbor.
2217 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2218 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2219 new->br_blockcount + right.br_blockcount,
2221 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2222 ifp->if_lastex = idx;
2224 rval = XFS_ILOG_FEXT(whichfork);
2227 if ((error = xfs_bmbt_lookup_eq(cur,
2229 right.br_startblock,
2230 right.br_blockcount, &i)))
2233 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2235 new->br_blockcount +
2236 right.br_blockcount,
2240 /* DELTA: One in-core extent grew. */
2241 temp = new->br_startoff;
2242 temp2 = new->br_blockcount +
2243 right.br_blockcount;
2248 * New allocation is not contiguous with another
2250 * Insert a new entry.
2252 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2254 xfs_iext_insert(ifp, idx, 1, new);
2255 ifp->if_lastex = idx;
2256 XFS_IFORK_NEXT_SET(ip, whichfork,
2257 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2259 rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2261 rval = XFS_ILOG_CORE;
2262 if ((error = xfs_bmbt_lookup_eq(cur,
2265 new->br_blockcount, &i)))
2268 cur->bc_rec.b.br_state = new->br_state;
2269 if ((error = xfs_bmbt_insert(cur, &i)))
2273 /* DELTA: A new extent was added in a hole. */
2274 temp = new->br_startoff;
2275 temp2 = new->br_blockcount;
2280 if (delta->xed_startoff > temp)
2281 delta->xed_startoff = temp;
2282 if (delta->xed_blockcount < temp2)
2283 delta->xed_blockcount = temp2;
2292 #undef STATE_SET_TEST
2297 * Adjust the size of the new extent based on di_extsize and rt extsize.
2300 xfs_bmap_extsize_align(
2302 xfs_bmbt_irec_t *gotp, /* next extent pointer */
2303 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2304 xfs_extlen_t extsz, /* align to this extent size */
2305 int rt, /* is this a realtime inode? */
2306 int eof, /* is extent at end-of-file? */
2307 int delay, /* creating delalloc extent? */
2308 int convert, /* overwriting unwritten extent? */
2309 xfs_fileoff_t *offp, /* in/out: aligned offset */
2310 xfs_extlen_t *lenp) /* in/out: aligned length */
2312 xfs_fileoff_t orig_off; /* original offset */
2313 xfs_extlen_t orig_alen; /* original length */
2314 xfs_fileoff_t orig_end; /* original off+len */
2315 xfs_fileoff_t nexto; /* next file offset */
2316 xfs_fileoff_t prevo; /* previous file offset */
2317 xfs_fileoff_t align_off; /* temp for offset */
2318 xfs_extlen_t align_alen; /* temp for length */
2319 xfs_extlen_t temp; /* temp for calculations */
2324 orig_off = align_off = *offp;
2325 orig_alen = align_alen = *lenp;
2326 orig_end = orig_off + orig_alen;
2329 * If this request overlaps an existing extent, then don't
2330 * attempt to perform any additional alignment.
2332 if (!delay && !eof &&
2333 (orig_off >= gotp->br_startoff) &&
2334 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2339 * If the file offset is unaligned vs. the extent size
2340 * we need to align it. This will be possible unless
2341 * the file was previously written with a kernel that didn't
2342 * perform this alignment, or if a truncate shot us in the
2345 temp = do_mod(orig_off, extsz);
2351 * Same adjustment for the end of the requested area.
2353 if ((temp = (align_alen % extsz))) {
2354 align_alen += extsz - temp;
2357 * If the previous block overlaps with this proposed allocation
2358 * then move the start forward without adjusting the length.
2360 if (prevp->br_startoff != NULLFILEOFF) {
2361 if (prevp->br_startblock == HOLESTARTBLOCK)
2362 prevo = prevp->br_startoff;
2364 prevo = prevp->br_startoff + prevp->br_blockcount;
2367 if (align_off != orig_off && align_off < prevo)
2370 * If the next block overlaps with this proposed allocation
2371 * then move the start back without adjusting the length,
2372 * but not before offset 0.
2373 * This may of course make the start overlap previous block,
2374 * and if we hit the offset 0 limit then the next block
2375 * can still overlap too.
2377 if (!eof && gotp->br_startoff != NULLFILEOFF) {
2378 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2379 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2380 nexto = gotp->br_startoff + gotp->br_blockcount;
2382 nexto = gotp->br_startoff;
2384 nexto = NULLFILEOFF;
2386 align_off + align_alen != orig_end &&
2387 align_off + align_alen > nexto)
2388 align_off = nexto > align_alen ? nexto - align_alen : 0;
2390 * If we're now overlapping the next or previous extent that
2391 * means we can't fit an extsz piece in this hole. Just move
2392 * the start forward to the first valid spot and set
2393 * the length so we hit the end.
2395 if (align_off != orig_off && align_off < prevo)
2397 if (align_off + align_alen != orig_end &&
2398 align_off + align_alen > nexto &&
2399 nexto != NULLFILEOFF) {
2400 ASSERT(nexto > prevo);
2401 align_alen = nexto - align_off;
2405 * If realtime, and the result isn't a multiple of the realtime
2406 * extent size we need to remove blocks until it is.
2408 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2410 * We're not covering the original request, or
2411 * we won't be able to once we fix the length.
2413 if (orig_off < align_off ||
2414 orig_end > align_off + align_alen ||
2415 align_alen - temp < orig_alen)
2416 return XFS_ERROR(EINVAL);
2418 * Try to fix it by moving the start up.
2420 if (align_off + temp <= orig_off) {
2425 * Try to fix it by moving the end in.
2427 else if (align_off + align_alen - temp >= orig_end)
2430 * Set the start to the minimum then trim the length.
2433 align_alen -= orig_off - align_off;
2434 align_off = orig_off;
2435 align_alen -= align_alen % mp->m_sb.sb_rextsize;
2438 * Result doesn't cover the request, fail it.
2440 if (orig_off < align_off || orig_end > align_off + align_alen)
2441 return XFS_ERROR(EINVAL);
2443 ASSERT(orig_off >= align_off);
2444 ASSERT(orig_end <= align_off + align_alen);
2448 if (!eof && gotp->br_startoff != NULLFILEOFF)
2449 ASSERT(align_off + align_alen <= gotp->br_startoff);
2450 if (prevp->br_startoff != NULLFILEOFF)
2451 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2459 #define XFS_ALLOC_GAP_UNITS 4
2463 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2465 xfs_fsblock_t adjust; /* adjustment to block numbers */
2466 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2467 xfs_mount_t *mp; /* mount point structure */
2468 int nullfb; /* true if ap->firstblock isn't set */
2469 int rt; /* true if inode is realtime */
2471 #define ISVALID(x,y) \
2473 (x) < mp->m_sb.sb_rblocks : \
2474 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2475 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2476 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2478 mp = ap->ip->i_mount;
2479 nullfb = ap->firstblock == NULLFSBLOCK;
2480 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2481 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2483 * If allocating at eof, and there's a previous real block,
2484 * try to use it's last block as our starting point.
2486 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2487 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2488 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2489 ap->prevp->br_startblock)) {
2490 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2492 * Adjust for the gap between prevp and us.
2495 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2497 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2501 * If not at eof, then compare the two neighbor blocks.
2502 * Figure out whether either one gives us a good starting point,
2503 * and pick the better one.
2505 else if (!ap->eof) {
2506 xfs_fsblock_t gotbno; /* right side block number */
2507 xfs_fsblock_t gotdiff=0; /* right side difference */
2508 xfs_fsblock_t prevbno; /* left side block number */
2509 xfs_fsblock_t prevdiff=0; /* left side difference */
2512 * If there's a previous (left) block, select a requested
2513 * start block based on it.
2515 if (ap->prevp->br_startoff != NULLFILEOFF &&
2516 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2517 (prevbno = ap->prevp->br_startblock +
2518 ap->prevp->br_blockcount) &&
2519 ISVALID(prevbno, ap->prevp->br_startblock)) {
2521 * Calculate gap to end of previous block.
2523 adjust = prevdiff = ap->off -
2524 (ap->prevp->br_startoff +
2525 ap->prevp->br_blockcount);
2527 * Figure the startblock based on the previous block's
2528 * end and the gap size.
2530 * If the gap is large relative to the piece we're
2531 * allocating, or using it gives us an invalid block
2532 * number, then just use the end of the previous block.
2534 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2535 ISVALID(prevbno + prevdiff,
2536 ap->prevp->br_startblock))
2541 * If the firstblock forbids it, can't use it,
2544 if (!rt && !nullfb &&
2545 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2546 prevbno = NULLFSBLOCK;
2549 * No previous block or can't follow it, just default.
2552 prevbno = NULLFSBLOCK;
2554 * If there's a following (right) block, select a requested
2555 * start block based on it.
2557 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2559 * Calculate gap to start of next block.
2561 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2563 * Figure the startblock based on the next block's
2564 * start and the gap size.
2566 gotbno = ap->gotp->br_startblock;
2569 * If the gap is large relative to the piece we're
2570 * allocating, or using it gives us an invalid block
2571 * number, then just use the start of the next block
2572 * offset by our length.
2574 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2575 ISVALID(gotbno - gotdiff, gotbno))
2577 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2579 gotdiff += adjust - ap->alen;
2583 * If the firstblock forbids it, can't use it,
2586 if (!rt && !nullfb &&
2587 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2588 gotbno = NULLFSBLOCK;
2591 * No next block, just default.
2594 gotbno = NULLFSBLOCK;
2596 * If both valid, pick the better one, else the only good
2597 * one, else ap->rval is already set (to 0 or the inode block).
2599 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2600 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2601 else if (prevbno != NULLFSBLOCK)
2603 else if (gotbno != NULLFSBLOCK)
2612 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2614 xfs_alloctype_t atype = 0; /* type for allocation routines */
2615 int error; /* error return value */
2616 xfs_mount_t *mp; /* mount point structure */
2617 xfs_extlen_t prod = 0; /* product factor for allocators */
2618 xfs_extlen_t ralen = 0; /* realtime allocation length */
2619 xfs_extlen_t align; /* minimum allocation alignment */
2620 xfs_rtblock_t rtx; /* realtime extent number */
2623 mp = ap->ip->i_mount;
2624 align = ap->ip->i_d.di_extsize ?
2625 ap->ip->i_d.di_extsize : mp->m_sb.sb_rextsize;
2626 prod = align / mp->m_sb.sb_rextsize;
2627 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2628 align, 1, ap->eof, 0,
2629 ap->conv, &ap->off, &ap->alen);
2633 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2636 * If the offset & length are not perfectly aligned
2637 * then kill prod, it will just get us in trouble.
2639 if (do_mod(ap->off, align) || ap->alen % align)
2642 * Set ralen to be the actual requested length in rtextents.
2644 ralen = ap->alen / mp->m_sb.sb_rextsize;
2646 * If the old value was close enough to MAXEXTLEN that
2647 * we rounded up to it, cut it back so it's valid again.
2648 * Note that if it's a really large request (bigger than
2649 * MAXEXTLEN), we don't hear about that number, and can't
2650 * adjust the starting point to match it.
2652 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2653 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2655 * If it's an allocation to an empty file at offset 0,
2656 * pick an extent that will space things out in the rt area.
2658 if (ap->eof && ap->off == 0) {
2659 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2662 ap->rval = rtx * mp->m_sb.sb_rextsize;
2667 xfs_bmap_adjacent(ap);
2670 * Realtime allocation, done through xfs_rtallocate_extent.
2672 atype = ap->rval == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2673 do_div(ap->rval, mp->m_sb.sb_rextsize);
2676 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2677 &ralen, atype, ap->wasdel, prod, &rtb)))
2679 if (rtb == NULLFSBLOCK && prod > 1 &&
2680 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2681 ap->alen, &ralen, atype,
2682 ap->wasdel, 1, &rtb)))
2685 if (ap->rval != NULLFSBLOCK) {
2686 ap->rval *= mp->m_sb.sb_rextsize;
2687 ralen *= mp->m_sb.sb_rextsize;
2689 ap->ip->i_d.di_nblocks += ralen;
2690 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2692 ap->ip->i_delayed_blks -= ralen;
2694 * Adjust the disk quota also. This was reserved
2697 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2698 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2699 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2708 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2710 xfs_mount_t *mp; /* mount point structure */
2711 xfs_alloctype_t atype = 0; /* type for allocation routines */
2712 xfs_extlen_t align; /* minimum allocation alignment */
2714 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2715 xfs_agnumber_t startag;
2716 xfs_alloc_arg_t args;
2719 xfs_extlen_t longest;
2721 xfs_extlen_t nextminlen = 0;
2723 int nullfb; /* true if ap->firstblock isn't set */
2729 mp = ap->ip->i_mount;
2730 align = (ap->userdata && ap->ip->i_d.di_extsize &&
2731 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)) ?
2732 ap->ip->i_d.di_extsize : 0;
2733 if (unlikely(align)) {
2734 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2735 align, 0, ap->eof, 0, ap->conv,
2736 &ap->off, &ap->alen);
2740 nullfb = ap->firstblock == NULLFSBLOCK;
2741 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2743 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2745 ap->rval = ap->firstblock;
2747 xfs_bmap_adjacent(ap);
2750 * If allowed, use ap->rval; otherwise must use firstblock since
2751 * it's in the right allocation group.
2753 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2756 ap->rval = ap->firstblock;
2758 * Normal allocation, done through xfs_alloc_vextent.
2760 tryagain = isaligned = 0;
2763 args.fsbno = ap->rval;
2764 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2767 args.type = XFS_ALLOCTYPE_START_BNO;
2768 args.total = ap->total;
2770 * Find the longest available space.
2771 * We're going to try for the whole allocation at once.
2773 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2775 down_read(&mp->m_peraglock);
2776 while (blen < ap->alen) {
2777 pag = &mp->m_perag[ag];
2778 if (!pag->pagf_init &&
2779 (error = xfs_alloc_pagf_init(mp, args.tp,
2780 ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2781 up_read(&mp->m_peraglock);
2785 * See xfs_alloc_fix_freelist...
2787 if (pag->pagf_init) {
2788 need = XFS_MIN_FREELIST_PAG(pag, mp);
2789 delta = need > pag->pagf_flcount ?
2790 need - pag->pagf_flcount : 0;
2791 longest = (pag->pagf_longest > delta) ?
2792 (pag->pagf_longest - delta) :
2793 (pag->pagf_flcount > 0 ||
2794 pag->pagf_longest > 0);
2799 if (++ag == mp->m_sb.sb_agcount)
2804 up_read(&mp->m_peraglock);
2806 * Since the above loop did a BUF_TRYLOCK, it is
2807 * possible that there is space for this request.
2809 if (notinit || blen < ap->minlen)
2810 args.minlen = ap->minlen;
2812 * If the best seen length is less than the request
2813 * length, use the best as the minimum.
2815 else if (blen < ap->alen)
2818 * Otherwise we've seen an extent as big as alen,
2819 * use that as the minimum.
2822 args.minlen = ap->alen;
2823 } else if (ap->low) {
2824 args.type = XFS_ALLOCTYPE_FIRST_AG;
2825 args.total = args.minlen = ap->minlen;
2827 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2828 args.total = ap->total;
2829 args.minlen = ap->minlen;
2831 if (unlikely(ap->userdata && ap->ip->i_d.di_extsize &&
2832 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE))) {
2833 args.prod = ap->ip->i_d.di_extsize;
2834 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2835 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2836 } else if (unlikely(mp->m_sb.sb_blocksize >= NBPP)) {
2840 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2841 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2842 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2845 * If we are not low on available data blocks, and the
2846 * underlying logical volume manager is a stripe, and
2847 * the file offset is zero then try to allocate data
2848 * blocks on stripe unit boundary.
2849 * NOTE: ap->aeof is only set if the allocation length
2850 * is >= the stripe unit and the allocation offset is
2851 * at the end of file.
2853 if (!ap->low && ap->aeof) {
2855 args.alignment = mp->m_dalign;
2859 * Adjust for alignment
2861 if (blen > args.alignment && blen <= ap->alen)
2862 args.minlen = blen - args.alignment;
2863 args.minalignslop = 0;
2866 * First try an exact bno allocation.
2867 * If it fails then do a near or start bno
2868 * allocation with alignment turned on.
2872 args.type = XFS_ALLOCTYPE_THIS_BNO;
2875 * Compute the minlen+alignment for the
2876 * next case. Set slop so that the value
2877 * of minlen+alignment+slop doesn't go up
2878 * between the calls.
2880 if (blen > mp->m_dalign && blen <= ap->alen)
2881 nextminlen = blen - mp->m_dalign;
2883 nextminlen = args.minlen;
2884 if (nextminlen + mp->m_dalign > args.minlen + 1)
2886 nextminlen + mp->m_dalign -
2889 args.minalignslop = 0;
2893 args.minalignslop = 0;
2895 args.minleft = ap->minleft;
2896 args.wasdel = ap->wasdel;
2898 args.userdata = ap->userdata;
2899 if ((error = xfs_alloc_vextent(&args)))
2901 if (tryagain && args.fsbno == NULLFSBLOCK) {
2903 * Exact allocation failed. Now try with alignment
2907 args.fsbno = ap->rval;
2908 args.alignment = mp->m_dalign;
2909 args.minlen = nextminlen;
2910 args.minalignslop = 0;
2912 if ((error = xfs_alloc_vextent(&args)))
2915 if (isaligned && args.fsbno == NULLFSBLOCK) {
2917 * allocation failed, so turn off alignment and
2921 args.fsbno = ap->rval;
2923 if ((error = xfs_alloc_vextent(&args)))
2926 if (args.fsbno == NULLFSBLOCK && nullfb &&
2927 args.minlen > ap->minlen) {
2928 args.minlen = ap->minlen;
2929 args.type = XFS_ALLOCTYPE_START_BNO;
2930 args.fsbno = ap->rval;
2931 if ((error = xfs_alloc_vextent(&args)))
2934 if (args.fsbno == NULLFSBLOCK && nullfb) {
2936 args.type = XFS_ALLOCTYPE_FIRST_AG;
2937 args.total = ap->minlen;
2939 if ((error = xfs_alloc_vextent(&args)))
2943 if (args.fsbno != NULLFSBLOCK) {
2944 ap->firstblock = ap->rval = args.fsbno;
2945 ASSERT(nullfb || fb_agno == args.agno ||
2946 (ap->low && fb_agno < args.agno));
2947 ap->alen = args.len;
2948 ap->ip->i_d.di_nblocks += args.len;
2949 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2951 ap->ip->i_delayed_blks -= args.len;
2953 * Adjust the disk quota also. This was reserved
2956 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2957 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2958 XFS_TRANS_DQ_BCOUNT,
2961 ap->rval = NULLFSBLOCK;
2968 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2969 * It figures out where to ask the underlying allocator to put the new extent.
2973 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2975 if ((ap->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && ap->userdata)
2976 return xfs_bmap_rtalloc(ap);
2977 return xfs_bmap_btalloc(ap);
2981 * Transform a btree format file with only one leaf node, where the
2982 * extents list will fit in the inode, into an extents format file.
2983 * Since the file extents are already in-core, all we have to do is
2984 * give up the space for the btree root and pitch the leaf block.
2986 STATIC int /* error */
2987 xfs_bmap_btree_to_extents(
2988 xfs_trans_t *tp, /* transaction pointer */
2989 xfs_inode_t *ip, /* incore inode pointer */
2990 xfs_btree_cur_t *cur, /* btree cursor */
2991 int *logflagsp, /* inode logging flags */
2992 int whichfork) /* data or attr fork */
2995 xfs_bmbt_block_t *cblock;/* child btree block */
2996 xfs_fsblock_t cbno; /* child block number */
2997 xfs_buf_t *cbp; /* child block's buffer */
2998 int error; /* error return value */
2999 xfs_ifork_t *ifp; /* inode fork data */
3000 xfs_mount_t *mp; /* mount point structure */
3001 xfs_bmbt_ptr_t *pp; /* ptr to block address */
3002 xfs_bmbt_block_t *rblock;/* root btree block */
3004 ifp = XFS_IFORK_PTR(ip, whichfork);
3005 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3006 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3007 rblock = ifp->if_broot;
3008 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3009 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3010 ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3012 pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3015 if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
3018 cbno = INT_GET(*pp, ARCH_CONVERT);
3019 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3020 XFS_BMAP_BTREE_REF)))
3022 cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3023 if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3025 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3026 ip->i_d.di_nblocks--;
3027 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3028 xfs_trans_binval(tp, cbp);
3029 if (cur->bc_bufs[0] == cbp)
3030 cur->bc_bufs[0] = NULL;
3031 xfs_iroot_realloc(ip, -1, whichfork);
3032 ASSERT(ifp->if_broot == NULL);
3033 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3034 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3035 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3040 * Called by xfs_bmapi to update file extent records and the btree
3041 * after removing space (or undoing a delayed allocation).
3043 STATIC int /* error */
3044 xfs_bmap_del_extent(
3045 xfs_inode_t *ip, /* incore inode pointer */
3046 xfs_trans_t *tp, /* current transaction pointer */
3047 xfs_extnum_t idx, /* extent number to update/delete */
3048 xfs_bmap_free_t *flist, /* list of extents to be freed */
3049 xfs_btree_cur_t *cur, /* if null, not a btree */
3050 xfs_bmbt_irec_t *del, /* data to remove from extents */
3051 int *logflagsp, /* inode logging flags */
3052 xfs_extdelta_t *delta, /* Change made to incore extents */
3053 int whichfork, /* data or attr fork */
3054 int rsvd) /* OK to allocate reserved blocks */
3056 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
3057 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
3058 xfs_fsblock_t del_endblock=0; /* first block past del */
3059 xfs_fileoff_t del_endoff; /* first offset past del */
3060 int delay; /* current block is delayed allocated */
3061 int do_fx; /* free extent at end of routine */
3062 xfs_bmbt_rec_t *ep; /* current extent entry pointer */
3063 int error; /* error return value */
3064 int flags; /* inode logging flags */
3065 #ifdef XFS_BMAP_TRACE
3066 static char fname[] = "xfs_bmap_del_extent";
3068 xfs_bmbt_irec_t got; /* current extent entry */
3069 xfs_fileoff_t got_endoff; /* first offset past got */
3070 int i; /* temp state */
3071 xfs_ifork_t *ifp; /* inode fork pointer */
3072 xfs_mount_t *mp; /* mount structure */
3073 xfs_filblks_t nblks; /* quota/sb block count */
3074 xfs_bmbt_irec_t new; /* new record to be inserted */
3076 uint qfield; /* quota field to update */
3077 xfs_filblks_t temp; /* for indirect length calculations */
3078 xfs_filblks_t temp2; /* for indirect length calculations */
3080 XFS_STATS_INC(xs_del_exlist);
3082 ifp = XFS_IFORK_PTR(ip, whichfork);
3083 ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3084 (uint)sizeof(xfs_bmbt_rec_t)));
3085 ASSERT(del->br_blockcount > 0);
3086 ep = xfs_iext_get_ext(ifp, idx);
3087 xfs_bmbt_get_all(ep, &got);
3088 ASSERT(got.br_startoff <= del->br_startoff);
3089 del_endoff = del->br_startoff + del->br_blockcount;
3090 got_endoff = got.br_startoff + got.br_blockcount;
3091 ASSERT(got_endoff >= del_endoff);
3092 delay = ISNULLSTARTBLOCK(got.br_startblock);
3093 ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3098 * If deleting a real allocation, must free up the disk space.
3101 flags = XFS_ILOG_CORE;
3103 * Realtime allocation. Free it and record di_nblocks update.
3105 if (whichfork == XFS_DATA_FORK &&
3106 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
3110 ASSERT(do_mod(del->br_blockcount,
3111 mp->m_sb.sb_rextsize) == 0);
3112 ASSERT(do_mod(del->br_startblock,
3113 mp->m_sb.sb_rextsize) == 0);
3114 bno = del->br_startblock;
3115 len = del->br_blockcount;
3116 do_div(bno, mp->m_sb.sb_rextsize);
3117 do_div(len, mp->m_sb.sb_rextsize);
3118 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3119 (xfs_extlen_t)len)))
3122 nblks = len * mp->m_sb.sb_rextsize;
3123 qfield = XFS_TRANS_DQ_RTBCOUNT;
3126 * Ordinary allocation.
3130 nblks = del->br_blockcount;
3131 qfield = XFS_TRANS_DQ_BCOUNT;
3134 * Set up del_endblock and cur for later.
3136 del_endblock = del->br_startblock + del->br_blockcount;
3138 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3139 got.br_startblock, got.br_blockcount,
3144 da_old = da_new = 0;
3146 da_old = STARTBLOCKVAL(got.br_startblock);
3152 * Set flag value to use in switch statement.
3153 * Left-contig is 2, right-contig is 1.
3155 switch (((got.br_startoff == del->br_startoff) << 1) |
3156 (got_endoff == del_endoff)) {
3159 * Matches the whole extent. Delete the entry.
3161 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
3162 xfs_iext_remove(ifp, idx, 1);
3163 ifp->if_lastex = idx;
3166 XFS_IFORK_NEXT_SET(ip, whichfork,
3167 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3168 flags |= XFS_ILOG_CORE;
3170 flags |= XFS_ILOG_FEXT(whichfork);
3173 if ((error = xfs_bmbt_delete(cur, &i)))
3180 * Deleting the first part of the extent.
3182 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
3183 xfs_bmbt_set_startoff(ep, del_endoff);
3184 temp = got.br_blockcount - del->br_blockcount;
3185 xfs_bmbt_set_blockcount(ep, temp);
3186 ifp->if_lastex = idx;
3188 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3190 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3191 xfs_bmap_trace_post_update(fname, "2", ip, idx,
3196 xfs_bmbt_set_startblock(ep, del_endblock);
3197 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
3199 flags |= XFS_ILOG_FEXT(whichfork);
3202 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3203 got.br_blockcount - del->br_blockcount,
3210 * Deleting the last part of the extent.
3212 temp = got.br_blockcount - del->br_blockcount;
3213 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
3214 xfs_bmbt_set_blockcount(ep, temp);
3215 ifp->if_lastex = idx;
3217 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3219 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3220 xfs_bmap_trace_post_update(fname, "1", ip, idx,
3225 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
3227 flags |= XFS_ILOG_FEXT(whichfork);
3230 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3232 got.br_blockcount - del->br_blockcount,
3239 * Deleting the middle of the extent.
3241 temp = del->br_startoff - got.br_startoff;
3242 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3243 xfs_bmbt_set_blockcount(ep, temp);
3244 new.br_startoff = del_endoff;
3245 temp2 = got_endoff - del_endoff;
3246 new.br_blockcount = temp2;
3247 new.br_state = got.br_state;
3249 new.br_startblock = del_endblock;
3250 flags |= XFS_ILOG_CORE;
3252 if ((error = xfs_bmbt_update(cur,
3254 got.br_startblock, temp,
3257 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3259 cur->bc_rec.b = new;
3260 error = xfs_bmbt_insert(cur, &i);
3261 if (error && error != ENOSPC)
3264 * If get no-space back from btree insert,
3265 * it tried a split, and we have a zero
3266 * block reservation.
3267 * Fix up our state and return the error.
3269 if (error == ENOSPC) {
3271 * Reset the cursor, don't trust
3272 * it after any insert operation.
3274 if ((error = xfs_bmbt_lookup_eq(cur,
3281 * Update the btree record back
3282 * to the original value.
3284 if ((error = xfs_bmbt_update(cur,
3291 * Reset the extent record back
3292 * to the original value.
3294 xfs_bmbt_set_blockcount(ep,
3297 error = XFS_ERROR(ENOSPC);
3302 flags |= XFS_ILOG_FEXT(whichfork);
3303 XFS_IFORK_NEXT_SET(ip, whichfork,
3304 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3306 ASSERT(whichfork == XFS_DATA_FORK);
3307 temp = xfs_bmap_worst_indlen(ip, temp);
3308 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3309 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3310 new.br_startblock = NULLSTARTBLOCK((int)temp2);
3311 da_new = temp + temp2;
3312 while (da_new > da_old) {
3316 xfs_bmbt_set_startblock(ep,
3317 NULLSTARTBLOCK((int)temp));
3319 if (da_new == da_old)
3325 NULLSTARTBLOCK((int)temp2);
3329 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3330 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3332 xfs_iext_insert(ifp, idx + 1, 1, &new);
3333 ifp->if_lastex = idx + 1;
3337 * If we need to, add to list of extents to delete.
3340 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3343 * Adjust inode # blocks in the file.
3346 ip->i_d.di_nblocks -= nblks;
3348 * Adjust quota data.
3351 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3354 * Account for change in delayed indirect blocks.
3355 * Nothing to do for disk quota accounting here.
3357 ASSERT(da_old >= da_new);
3358 if (da_old > da_new)
3359 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3362 /* DELTA: report the original extent. */
3363 if (delta->xed_startoff > got.br_startoff)
3364 delta->xed_startoff = got.br_startoff;
3365 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3366 delta->xed_blockcount = got.br_startoff +
3375 * Remove the entry "free" from the free item list. Prev points to the
3376 * previous entry, unless "free" is the head of the list.
3380 xfs_bmap_free_t *flist, /* free item list header */
3381 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3382 xfs_bmap_free_item_t *free) /* list item to be freed */
3385 prev->xbfi_next = free->xbfi_next;
3387 flist->xbf_first = free->xbfi_next;
3389 kmem_zone_free(xfs_bmap_free_item_zone, free);
3393 * Convert an extents-format file into a btree-format file.
3394 * The new file will have a root block (in the inode) and a single child block.
3396 STATIC int /* error */
3397 xfs_bmap_extents_to_btree(
3398 xfs_trans_t *tp, /* transaction pointer */
3399 xfs_inode_t *ip, /* incore inode pointer */
3400 xfs_fsblock_t *firstblock, /* first-block-allocated */
3401 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3402 xfs_btree_cur_t **curp, /* cursor returned to caller */
3403 int wasdel, /* converting a delayed alloc */
3404 int *logflagsp, /* inode logging flags */
3405 int whichfork) /* data or attr fork */
3407 xfs_bmbt_block_t *ablock; /* allocated (child) bt block */
3408 xfs_buf_t *abp; /* buffer for ablock */
3409 xfs_alloc_arg_t args; /* allocation arguments */
3410 xfs_bmbt_rec_t *arp; /* child record pointer */
3411 xfs_bmbt_block_t *block; /* btree root block */
3412 xfs_btree_cur_t *cur; /* bmap btree cursor */
3413 xfs_bmbt_rec_t *ep; /* extent record pointer */
3414 int error; /* error return value */
3415 xfs_extnum_t i, cnt; /* extent record index */
3416 xfs_ifork_t *ifp; /* inode fork pointer */
3417 xfs_bmbt_key_t *kp; /* root block key pointer */
3418 xfs_mount_t *mp; /* mount structure */
3419 xfs_extnum_t nextents; /* number of file extents */
3420 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3422 ifp = XFS_IFORK_PTR(ip, whichfork);
3423 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3424 ASSERT(ifp->if_ext_max ==
3425 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3427 * Make space in the inode incore.
3429 xfs_iroot_realloc(ip, 1, whichfork);
3430 ifp->if_flags |= XFS_IFBROOT;
3434 block = ifp->if_broot;
3435 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3436 block->bb_level = cpu_to_be16(1);
3437 block->bb_numrecs = cpu_to_be16(1);
3438 block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3439 block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3441 * Need a cursor. Can't allocate until bb_level is filled in.
3444 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3446 cur->bc_private.b.firstblock = *firstblock;
3447 cur->bc_private.b.flist = flist;
3448 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3450 * Convert to a btree with two levels, one record in root.
3452 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3455 if (*firstblock == NULLFSBLOCK) {
3456 args.type = XFS_ALLOCTYPE_START_BNO;
3457 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3458 } else if (flist->xbf_low) {
3459 args.type = XFS_ALLOCTYPE_START_BNO;
3460 args.fsbno = *firstblock;
3462 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3463 args.fsbno = *firstblock;
3465 args.minlen = args.maxlen = args.prod = 1;
3466 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3467 args.minalignslop = 0;
3468 args.wasdel = wasdel;
3470 if ((error = xfs_alloc_vextent(&args))) {
3471 xfs_iroot_realloc(ip, -1, whichfork);
3472 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3476 * Allocation can't fail, the space was reserved.
3478 ASSERT(args.fsbno != NULLFSBLOCK);
3479 ASSERT(*firstblock == NULLFSBLOCK ||
3480 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3482 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3483 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3484 cur->bc_private.b.allocated++;
3485 ip->i_d.di_nblocks++;
3486 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3487 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3489 * Fill in the child block.
3491 ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3492 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3493 ablock->bb_level = 0;
3494 ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3495 ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3496 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3497 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3498 for (cnt = i = 0; i < nextents; i++) {
3499 ep = xfs_iext_get_ext(ifp, i);
3500 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3501 arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3502 arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3506 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3507 ablock->bb_numrecs = cpu_to_be16(cnt);
3509 * Fill in the root key and pointer.
3511 kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3512 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3513 INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3514 pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3515 INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3517 * Do all this logging at the end so that
3518 * the root is at the right level.
3520 xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3521 xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3522 ASSERT(*curp == NULL);
3524 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3529 * Helper routine to reset inode di_forkoff field when switching
3530 * attribute fork from local to extent format - we reset it where
3531 * possible to make space available for inline data fork extents.
3534 xfs_bmap_forkoff_reset(
3539 if (whichfork == XFS_ATTR_FORK &&
3540 (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3541 (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3542 ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3543 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3544 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3545 (uint)sizeof(xfs_bmbt_rec_t);
3546 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3547 (uint)sizeof(xfs_bmbt_rec_t);
3552 * Convert a local file to an extents file.
3553 * This code is out of bounds for data forks of regular files,
3554 * since the file data needs to get logged so things will stay consistent.
3555 * (The bmap-level manipulations are ok, though).
3557 STATIC int /* error */
3558 xfs_bmap_local_to_extents(
3559 xfs_trans_t *tp, /* transaction pointer */
3560 xfs_inode_t *ip, /* incore inode pointer */
3561 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3562 xfs_extlen_t total, /* total blocks needed by transaction */
3563 int *logflagsp, /* inode logging flags */
3564 int whichfork) /* data or attr fork */
3566 int error; /* error return value */
3567 int flags; /* logging flags returned */
3568 #ifdef XFS_BMAP_TRACE
3569 static char fname[] = "xfs_bmap_local_to_extents";
3571 xfs_ifork_t *ifp; /* inode fork pointer */
3574 * We don't want to deal with the case of keeping inode data inline yet.
3575 * So sending the data fork of a regular inode is invalid.
3577 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3578 whichfork == XFS_DATA_FORK));
3579 ifp = XFS_IFORK_PTR(ip, whichfork);
3580 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3583 if (ifp->if_bytes) {
3584 xfs_alloc_arg_t args; /* allocation arguments */
3585 xfs_buf_t *bp; /* buffer for extent block */
3586 xfs_bmbt_rec_t *ep; /* extent record pointer */
3589 args.mp = ip->i_mount;
3590 ASSERT((ifp->if_flags &
3591 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3593 * Allocate a block. We know we need only one, since the
3594 * file currently fits in an inode.
3596 if (*firstblock == NULLFSBLOCK) {
3597 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3598 args.type = XFS_ALLOCTYPE_START_BNO;
3600 args.fsbno = *firstblock;
3601 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3604 args.mod = args.minleft = args.alignment = args.wasdel =
3605 args.isfl = args.minalignslop = 0;
3606 args.minlen = args.maxlen = args.prod = 1;
3607 if ((error = xfs_alloc_vextent(&args)))
3610 * Can't fail, the space was reserved.
3612 ASSERT(args.fsbno != NULLFSBLOCK);
3613 ASSERT(args.len == 1);
3614 *firstblock = args.fsbno;
3615 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3616 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3618 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3619 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3620 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3621 xfs_iext_add(ifp, 0, 1);
3622 ep = xfs_iext_get_ext(ifp, 0);
3623 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3624 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3625 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3626 ip->i_d.di_nblocks = 1;
3627 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3628 XFS_TRANS_DQ_BCOUNT, 1L);
3629 flags |= XFS_ILOG_FEXT(whichfork);
3631 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3632 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3634 ifp->if_flags &= ~XFS_IFINLINE;
3635 ifp->if_flags |= XFS_IFEXTENTS;
3636 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3637 flags |= XFS_ILOG_CORE;
3644 * Search the extent records for the entry containing block bno.
3645 * If bno lies in a hole, point to the next entry. If bno lies
3646 * past eof, *eofp will be set, and *prevp will contain the last
3647 * entry (null if none). Else, *lastxp will be set to the index
3648 * of the found entry; *gotp will contain the entry.
3650 xfs_bmbt_rec_t * /* pointer to found extent entry */
3651 xfs_bmap_search_multi_extents(
3652 xfs_ifork_t *ifp, /* inode fork pointer */
3653 xfs_fileoff_t bno, /* block number searched for */
3654 int *eofp, /* out: end of file found */
3655 xfs_extnum_t *lastxp, /* out: last extent index */
3656 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3657 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3659 xfs_bmbt_rec_t *ep; /* extent record pointer */
3660 xfs_extnum_t lastx; /* last extent index */
3663 * Initialize the extent entry structure to catch access to
3664 * uninitialized br_startblock field.
3666 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3667 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3668 gotp->br_state = XFS_EXT_INVALID;
3670 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3672 gotp->br_startblock = 0xffffa5a5;
3674 prevp->br_startoff = NULLFILEOFF;
3676 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3678 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3680 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3681 xfs_bmbt_get_all(ep, gotp);
3695 * Search the extents list for the inode, for the extent containing bno.
3696 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3697 * *eofp will be set, and *prevp will contain the last entry (null if none).
3698 * Else, *lastxp will be set to the index of the found
3699 * entry; *gotp will contain the entry.
3701 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
3702 xfs_bmap_search_extents(
3703 xfs_inode_t *ip, /* incore inode pointer */
3704 xfs_fileoff_t bno, /* block number searched for */
3705 int whichfork, /* data or attr fork */
3706 int *eofp, /* out: end of file found */
3707 xfs_extnum_t *lastxp, /* out: last extent index */
3708 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3709 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3711 xfs_ifork_t *ifp; /* inode fork pointer */
3712 xfs_bmbt_rec_t *ep; /* extent record pointer */
3713 int rt; /* realtime flag */
3715 XFS_STATS_INC(xs_look_exlist);
3716 ifp = XFS_IFORK_PTR(ip, whichfork);
3718 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3720 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
3721 if (unlikely(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM))) {
3722 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3723 "start_block : %llx start_off : %llx blkcnt : %llx "
3724 "extent-state : %x \n",
3725 (ip->i_mount)->m_fsname, (long long)ip->i_ino,
3726 (unsigned long long)gotp->br_startblock,
3727 (unsigned long long)gotp->br_startoff,
3728 (unsigned long long)gotp->br_blockcount,
3735 #ifdef XFS_BMAP_TRACE
3736 ktrace_t *xfs_bmap_trace_buf;
3739 * Add a bmap trace buffer entry. Base routine for the others.
3742 xfs_bmap_trace_addentry(
3743 int opcode, /* operation */
3744 char *fname, /* function name */
3745 char *desc, /* operation description */
3746 xfs_inode_t *ip, /* incore inode pointer */
3747 xfs_extnum_t idx, /* index of entry(ies) */
3748 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
3749 xfs_bmbt_rec_t *r1, /* first record */
3750 xfs_bmbt_rec_t *r2, /* second record or null */
3751 int whichfork) /* data or attr fork */
3755 ASSERT(cnt == 1 || cnt == 2);
3760 memset(&tr2, 0, sizeof(tr2));
3763 ktrace_enter(xfs_bmap_trace_buf,
3764 (void *)(__psint_t)(opcode | (whichfork << 16)),
3765 (void *)fname, (void *)desc, (void *)ip,
3766 (void *)(__psint_t)idx,
3767 (void *)(__psint_t)cnt,
3768 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3769 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3770 (void *)(__psunsigned_t)(r1->l0 >> 32),
3771 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3772 (void *)(__psunsigned_t)(r1->l1 >> 32),
3773 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3774 (void *)(__psunsigned_t)(r2->l0 >> 32),
3775 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3776 (void *)(__psunsigned_t)(r2->l1 >> 32),
3777 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3779 ASSERT(ip->i_xtrace);
3780 ktrace_enter(ip->i_xtrace,
3781 (void *)(__psint_t)(opcode | (whichfork << 16)),
3782 (void *)fname, (void *)desc, (void *)ip,
3783 (void *)(__psint_t)idx,
3784 (void *)(__psint_t)cnt,
3785 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3786 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3787 (void *)(__psunsigned_t)(r1->l0 >> 32),
3788 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3789 (void *)(__psunsigned_t)(r1->l1 >> 32),
3790 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3791 (void *)(__psunsigned_t)(r2->l0 >> 32),
3792 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3793 (void *)(__psunsigned_t)(r2->l1 >> 32),
3794 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3799 * Add bmap trace entry prior to a call to xfs_iext_remove.
3802 xfs_bmap_trace_delete(
3803 char *fname, /* function name */
3804 char *desc, /* operation description */
3805 xfs_inode_t *ip, /* incore inode pointer */
3806 xfs_extnum_t idx, /* index of entry(entries) deleted */
3807 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
3808 int whichfork) /* data or attr fork */
3810 xfs_ifork_t *ifp; /* inode fork pointer */
3812 ifp = XFS_IFORK_PTR(ip, whichfork);
3813 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3814 cnt, xfs_iext_get_ext(ifp, idx),
3815 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3820 * Add bmap trace entry prior to a call to xfs_iext_insert, or
3821 * reading in the extents list from the disk (in the btree).
3824 xfs_bmap_trace_insert(
3825 char *fname, /* function name */
3826 char *desc, /* operation description */
3827 xfs_inode_t *ip, /* incore inode pointer */
3828 xfs_extnum_t idx, /* index of entry(entries) inserted */
3829 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
3830 xfs_bmbt_irec_t *r1, /* inserted record 1 */
3831 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
3832 int whichfork) /* data or attr fork */
3834 xfs_bmbt_rec_t tr1; /* compressed record 1 */
3835 xfs_bmbt_rec_t tr2; /* compressed record 2 if needed */
3837 xfs_bmbt_set_all(&tr1, r1);
3840 xfs_bmbt_set_all(&tr2, r2);
3845 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3846 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3850 * Add bmap trace entry after updating an extent record in place.
3853 xfs_bmap_trace_post_update(
3854 char *fname, /* function name */
3855 char *desc, /* operation description */
3856 xfs_inode_t *ip, /* incore inode pointer */
3857 xfs_extnum_t idx, /* index of entry updated */
3858 int whichfork) /* data or attr fork */
3860 xfs_ifork_t *ifp; /* inode fork pointer */
3862 ifp = XFS_IFORK_PTR(ip, whichfork);
3863 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3864 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3868 * Add bmap trace entry prior to updating an extent record in place.
3871 xfs_bmap_trace_pre_update(
3872 char *fname, /* function name */
3873 char *desc, /* operation description */
3874 xfs_inode_t *ip, /* incore inode pointer */
3875 xfs_extnum_t idx, /* index of entry to be updated */
3876 int whichfork) /* data or attr fork */
3878 xfs_ifork_t *ifp; /* inode fork pointer */
3880 ifp = XFS_IFORK_PTR(ip, whichfork);
3881 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3882 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3884 #endif /* XFS_BMAP_TRACE */
3887 * Compute the worst-case number of indirect blocks that will be used
3888 * for ip's delayed extent of length "len".
3890 STATIC xfs_filblks_t
3891 xfs_bmap_worst_indlen(
3892 xfs_inode_t *ip, /* incore inode pointer */
3893 xfs_filblks_t len) /* delayed extent length */
3895 int level; /* btree level number */
3896 int maxrecs; /* maximum record count at this level */
3897 xfs_mount_t *mp; /* mount structure */
3898 xfs_filblks_t rval; /* return value */
3901 maxrecs = mp->m_bmap_dmxr[0];
3902 for (level = 0, rval = 0;
3903 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3906 do_div(len, maxrecs);
3909 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3912 maxrecs = mp->m_bmap_dmxr[1];
3917 #if defined(XFS_RW_TRACE)
3926 if (ip->i_rwtrace == NULL)
3928 ktrace_enter(ip->i_rwtrace,
3929 (void *)(__psint_t)XFS_BUNMAP,
3931 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3932 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3933 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3934 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3935 (void *)(__psint_t)len,
3936 (void *)(__psint_t)flags,
3937 (void *)(unsigned long)current_cpu(),
3949 * Convert inode from non-attributed to attributed.
3950 * Must not be in a transaction, ip must not be locked.
3952 int /* error code */
3953 xfs_bmap_add_attrfork(
3954 xfs_inode_t *ip, /* incore inode pointer */
3955 int size, /* space new attribute needs */
3956 int rsvd) /* xact may use reserved blks */
3958 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3959 xfs_bmap_free_t flist; /* freed extent records */
3960 xfs_mount_t *mp; /* mount structure */
3961 xfs_trans_t *tp; /* transaction pointer */
3962 unsigned long s; /* spinlock spl value */
3963 int blks; /* space reservation */
3964 int version = 1; /* superblock attr version */
3965 int committed; /* xaction was committed */
3966 int logflags; /* logging flags */
3967 int error; /* error return value */
3969 ASSERT(XFS_IFORK_Q(ip) == 0);
3970 ASSERT(ip->i_df.if_ext_max ==
3971 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3974 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3975 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3976 blks = XFS_ADDAFORK_SPACE_RES(mp);
3978 tp->t_flags |= XFS_TRANS_RESERVE;
3979 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3980 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3982 xfs_ilock(ip, XFS_ILOCK_EXCL);
3983 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3984 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3985 XFS_QMOPT_RES_REGBLKS);
3987 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3988 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3991 if (XFS_IFORK_Q(ip))
3993 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3995 * For inodes coming from pre-6.2 filesystems.
3997 ASSERT(ip->i_d.di_aformat == 0);
3998 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
4000 ASSERT(ip->i_d.di_anextents == 0);
4001 VN_HOLD(XFS_ITOV(ip));
4002 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4003 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4004 switch (ip->i_d.di_format) {
4005 case XFS_DINODE_FMT_DEV:
4006 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4008 case XFS_DINODE_FMT_UUID:
4009 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4011 case XFS_DINODE_FMT_LOCAL:
4012 case XFS_DINODE_FMT_EXTENTS:
4013 case XFS_DINODE_FMT_BTREE:
4014 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4015 if (!ip->i_d.di_forkoff)
4016 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4017 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4022 error = XFS_ERROR(EINVAL);
4025 ip->i_df.if_ext_max =
4026 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4027 ASSERT(ip->i_afp == NULL);
4028 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4029 ip->i_afp->if_ext_max =
4030 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4031 ip->i_afp->if_flags = XFS_IFEXTENTS;
4033 XFS_BMAP_INIT(&flist, &firstblock);
4034 switch (ip->i_d.di_format) {
4035 case XFS_DINODE_FMT_LOCAL:
4036 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4039 case XFS_DINODE_FMT_EXTENTS:
4040 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4043 case XFS_DINODE_FMT_BTREE:
4044 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4052 xfs_trans_log_inode(tp, ip, logflags);
4055 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
4056 (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
4057 __int64_t sbfields = 0;
4059 s = XFS_SB_LOCK(mp);
4060 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
4061 XFS_SB_VERSION_ADDATTR(&mp->m_sb);
4062 sbfields |= XFS_SB_VERSIONNUM;
4064 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
4065 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
4066 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4069 XFS_SB_UNLOCK(mp, s);
4070 xfs_mod_sb(tp, sbfields);
4072 XFS_SB_UNLOCK(mp, s);
4074 if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
4076 error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
4077 ASSERT(ip->i_df.if_ext_max ==
4078 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4081 xfs_bmap_cancel(&flist);
4083 ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
4084 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4086 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4087 ASSERT(ip->i_df.if_ext_max ==
4088 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4093 * Add the extent to the list of extents to be free at transaction end.
4094 * The list is maintained sorted (by block number).
4099 xfs_fsblock_t bno, /* fs block number of extent */
4100 xfs_filblks_t len, /* length of extent */
4101 xfs_bmap_free_t *flist, /* list of extents */
4102 xfs_mount_t *mp) /* mount point structure */
4104 xfs_bmap_free_item_t *cur; /* current (next) element */
4105 xfs_bmap_free_item_t *new; /* new element */
4106 xfs_bmap_free_item_t *prev; /* previous element */
4108 xfs_agnumber_t agno;
4109 xfs_agblock_t agbno;
4111 ASSERT(bno != NULLFSBLOCK);
4113 ASSERT(len <= MAXEXTLEN);
4114 ASSERT(!ISNULLSTARTBLOCK(bno));
4115 agno = XFS_FSB_TO_AGNO(mp, bno);
4116 agbno = XFS_FSB_TO_AGBNO(mp, bno);
4117 ASSERT(agno < mp->m_sb.sb_agcount);
4118 ASSERT(agbno < mp->m_sb.sb_agblocks);
4119 ASSERT(len < mp->m_sb.sb_agblocks);
4120 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4122 ASSERT(xfs_bmap_free_item_zone != NULL);
4123 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4124 new->xbfi_startblock = bno;
4125 new->xbfi_blockcount = (xfs_extlen_t)len;
4126 for (prev = NULL, cur = flist->xbf_first;
4128 prev = cur, cur = cur->xbfi_next) {
4129 if (cur->xbfi_startblock >= bno)
4133 prev->xbfi_next = new;
4135 flist->xbf_first = new;
4136 new->xbfi_next = cur;
4141 * Compute and fill in the value of the maximum depth of a bmap btree
4142 * in this filesystem. Done once, during mount.
4145 xfs_bmap_compute_maxlevels(
4146 xfs_mount_t *mp, /* file system mount structure */
4147 int whichfork) /* data or attr fork */
4149 int level; /* btree level */
4150 uint maxblocks; /* max blocks at this level */
4151 uint maxleafents; /* max leaf entries possible */
4152 int maxrootrecs; /* max records in root block */
4153 int minleafrecs; /* min records in leaf block */
4154 int minnoderecs; /* min records in node block */
4155 int sz; /* root block size */
4158 * The maximum number of extents in a file, hence the maximum
4159 * number of leaf entries, is controlled by the type of di_nextents
4160 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4161 * (a signed 16-bit number, xfs_aextnum_t).
4163 if (whichfork == XFS_DATA_FORK) {
4164 maxleafents = MAXEXTNUM;
4165 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4166 XFS_BMDR_SPACE_CALC(MINDBTPTRS) : mp->m_attroffset;
4168 maxleafents = MAXAEXTNUM;
4169 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4170 XFS_BMDR_SPACE_CALC(MINABTPTRS) :
4171 mp->m_sb.sb_inodesize - mp->m_attroffset;
4173 maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4174 minleafrecs = mp->m_bmap_dmnr[0];
4175 minnoderecs = mp->m_bmap_dmnr[1];
4176 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4177 for (level = 1; maxblocks > 1; level++) {
4178 if (maxblocks <= maxrootrecs)
4181 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4183 mp->m_bm_maxlevels[whichfork] = level;
4187 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4188 * caller. Frees all the extents that need freeing, which must be done
4189 * last due to locking considerations. We never free any extents in
4190 * the first transaction. This is to allow the caller to make the first
4191 * transaction a synchronous one so that the pointers to the data being
4192 * broken in this transaction will be permanent before the data is actually
4193 * freed. This is necessary to prevent blocks from being reallocated
4194 * and written to before the free and reallocation are actually permanent.
4195 * We do not just make the first transaction synchronous here, because
4196 * there are more efficient ways to gain the same protection in some cases
4197 * (see the file truncation code).
4199 * Return 1 if the given transaction was committed and a new one
4200 * started, and 0 otherwise in the committed parameter.
4205 xfs_trans_t **tp, /* transaction pointer addr */
4206 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4207 xfs_fsblock_t firstblock, /* controlled ag for allocs */
4208 int *committed) /* xact committed or not */
4210 xfs_efd_log_item_t *efd; /* extent free data */
4211 xfs_efi_log_item_t *efi; /* extent free intention */
4212 int error; /* error return value */
4213 xfs_bmap_free_item_t *free; /* free extent item */
4214 unsigned int logres; /* new log reservation */
4215 unsigned int logcount; /* new log count */
4216 xfs_mount_t *mp; /* filesystem mount structure */
4217 xfs_bmap_free_item_t *next; /* next item on free list */
4218 xfs_trans_t *ntp; /* new transaction pointer */
4220 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4221 if (flist->xbf_count == 0) {
4226 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4227 for (free = flist->xbf_first; free; free = free->xbfi_next)
4228 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4229 free->xbfi_blockcount);
4230 logres = ntp->t_log_res;
4231 logcount = ntp->t_log_count;
4232 ntp = xfs_trans_dup(*tp);
4233 error = xfs_trans_commit(*tp, 0, NULL);
4237 * We have a new transaction, so we should return committed=1,
4238 * even though we're returning an error.
4243 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4246 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4247 for (free = flist->xbf_first; free != NULL; free = next) {
4248 next = free->xbfi_next;
4249 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4250 free->xbfi_blockcount))) {
4252 * The bmap free list will be cleaned up at a
4253 * higher level. The EFI will be canceled when
4254 * this transaction is aborted.
4255 * Need to force shutdown here to make sure it
4256 * happens, since this transaction may not be
4260 if (!XFS_FORCED_SHUTDOWN(mp))
4261 xfs_force_shutdown(mp,
4262 (error == EFSCORRUPTED) ?
4263 XFS_CORRUPT_INCORE :
4264 XFS_METADATA_IO_ERROR);
4267 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4268 free->xbfi_blockcount);
4269 xfs_bmap_del_free(flist, NULL, free);
4275 * Free up any items left in the list.
4279 xfs_bmap_free_t *flist) /* list of bmap_free_items */
4281 xfs_bmap_free_item_t *free; /* free list item */
4282 xfs_bmap_free_item_t *next;
4284 if (flist->xbf_count == 0)
4286 ASSERT(flist->xbf_first != NULL);
4287 for (free = flist->xbf_first; free; free = next) {
4288 next = free->xbfi_next;
4289 xfs_bmap_del_free(flist, NULL, free);
4291 ASSERT(flist->xbf_count == 0);
4295 * Returns the file-relative block number of the first unused block(s)
4296 * in the file with at least "len" logically contiguous blocks free.
4297 * This is the lowest-address hole if the file has holes, else the first block
4298 * past the end of file.
4299 * Return 0 if the file is currently local (in-inode).
4302 xfs_bmap_first_unused(
4303 xfs_trans_t *tp, /* transaction pointer */
4304 xfs_inode_t *ip, /* incore inode */
4305 xfs_extlen_t len, /* size of hole to find */
4306 xfs_fileoff_t *first_unused, /* unused block */
4307 int whichfork) /* data or attr fork */
4309 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
4310 int error; /* error return value */
4311 int idx; /* extent record index */
4312 xfs_ifork_t *ifp; /* inode fork pointer */
4313 xfs_fileoff_t lastaddr; /* last block number seen */
4314 xfs_fileoff_t lowest; /* lowest useful block */
4315 xfs_fileoff_t max; /* starting useful block */
4316 xfs_fileoff_t off; /* offset for this block */
4317 xfs_extnum_t nextents; /* number of extent entries */
4319 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4320 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4321 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4322 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4326 ifp = XFS_IFORK_PTR(ip, whichfork);
4327 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4328 (error = xfs_iread_extents(tp, ip, whichfork)))
4330 lowest = *first_unused;
4331 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4332 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4333 ep = xfs_iext_get_ext(ifp, idx);
4334 off = xfs_bmbt_get_startoff(ep);
4336 * See if the hole before this extent will work.
4338 if (off >= lowest + len && off - max >= len) {
4339 *first_unused = max;
4342 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4343 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4345 *first_unused = max;
4350 * Returns the file-relative block number of the last block + 1 before
4351 * last_block (input value) in the file.
4352 * This is not based on i_size, it is based on the extent records.
4353 * Returns 0 for local files, as they do not have extent records.
4356 xfs_bmap_last_before(
4357 xfs_trans_t *tp, /* transaction pointer */
4358 xfs_inode_t *ip, /* incore inode */
4359 xfs_fileoff_t *last_block, /* last block */
4360 int whichfork) /* data or attr fork */
4362 xfs_fileoff_t bno; /* input file offset */
4363 int eof; /* hit end of file */
4364 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4365 int error; /* error return value */
4366 xfs_bmbt_irec_t got; /* current extent value */
4367 xfs_ifork_t *ifp; /* inode fork pointer */
4368 xfs_extnum_t lastx; /* last extent used */
4369 xfs_bmbt_irec_t prev; /* previous extent value */
4371 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4372 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4373 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4374 return XFS_ERROR(EIO);
4375 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4379 ifp = XFS_IFORK_PTR(ip, whichfork);
4380 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4381 (error = xfs_iread_extents(tp, ip, whichfork)))
4383 bno = *last_block - 1;
4384 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4386 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4387 if (prev.br_startoff == NULLFILEOFF)
4390 *last_block = prev.br_startoff + prev.br_blockcount;
4393 * Otherwise *last_block is already the right answer.
4399 * Returns the file-relative block number of the first block past eof in
4400 * the file. This is not based on i_size, it is based on the extent records.
4401 * Returns 0 for local files, as they do not have extent records.
4404 xfs_bmap_last_offset(
4405 xfs_trans_t *tp, /* transaction pointer */
4406 xfs_inode_t *ip, /* incore inode */
4407 xfs_fileoff_t *last_block, /* last block */
4408 int whichfork) /* data or attr fork */
4410 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4411 int error; /* error return value */
4412 xfs_ifork_t *ifp; /* inode fork pointer */
4413 xfs_extnum_t nextents; /* number of extent entries */
4415 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4416 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4417 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4418 return XFS_ERROR(EIO);
4419 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4423 ifp = XFS_IFORK_PTR(ip, whichfork);
4424 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4425 (error = xfs_iread_extents(tp, ip, whichfork)))
4427 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4432 ep = xfs_iext_get_ext(ifp, nextents - 1);
4433 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4438 * Returns whether the selected fork of the inode has exactly one
4439 * block or not. For the data fork we check this matches di_size,
4440 * implying the file's range is 0..bsize-1.
4442 int /* 1=>1 block, 0=>otherwise */
4444 xfs_inode_t *ip, /* incore inode */
4445 int whichfork) /* data or attr fork */
4447 xfs_bmbt_rec_t *ep; /* ptr to fork's extent */
4448 xfs_ifork_t *ifp; /* inode fork pointer */
4449 int rval; /* return value */
4450 xfs_bmbt_irec_t s; /* internal version of extent */
4453 if (whichfork == XFS_DATA_FORK)
4454 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4456 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4458 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4460 ifp = XFS_IFORK_PTR(ip, whichfork);
4461 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4462 ep = xfs_iext_get_ext(ifp, 0);
4463 xfs_bmbt_get_all(ep, &s);
4464 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4465 if (rval && whichfork == XFS_DATA_FORK)
4466 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4471 * Read in the extents to if_extents.
4472 * All inode fields are set up by caller, we just traverse the btree
4473 * and copy the records in. If the file system cannot contain unwritten
4474 * extents, the records are checked for no "state" flags.
4477 xfs_bmap_read_extents(
4478 xfs_trans_t *tp, /* transaction pointer */
4479 xfs_inode_t *ip, /* incore inode */
4480 int whichfork) /* data or attr fork */
4482 xfs_bmbt_block_t *block; /* current btree block */
4483 xfs_fsblock_t bno; /* block # of "block" */
4484 xfs_buf_t *bp; /* buffer for "block" */
4485 int error; /* error return value */
4486 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4487 #ifdef XFS_BMAP_TRACE
4488 static char fname[] = "xfs_bmap_read_extents";
4490 xfs_extnum_t i, j; /* index into the extents list */
4491 xfs_ifork_t *ifp; /* fork structure */
4492 int level; /* btree level, for checking */
4493 xfs_mount_t *mp; /* file system mount structure */
4494 xfs_bmbt_ptr_t *pp; /* pointer to block address */
4496 xfs_extnum_t room; /* number of entries there's room for */
4500 ifp = XFS_IFORK_PTR(ip, whichfork);
4501 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4502 XFS_EXTFMT_INODE(ip);
4503 block = ifp->if_broot;
4505 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4507 level = be16_to_cpu(block->bb_level);
4509 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4510 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4511 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4512 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4513 bno = INT_GET(*pp, ARCH_CONVERT);
4515 * Go down the tree until leaf level is reached, following the first
4516 * pointer (leftmost) at each level.
4518 while (level-- > 0) {
4519 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4520 XFS_BMAP_BTREE_REF)))
4522 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4523 XFS_WANT_CORRUPTED_GOTO(
4524 XFS_BMAP_SANITY_CHECK(mp, block, level),
4528 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4529 1, mp->m_bmap_dmxr[1]);
4530 XFS_WANT_CORRUPTED_GOTO(
4531 XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4533 bno = INT_GET(*pp, ARCH_CONVERT);
4534 xfs_trans_brelse(tp, bp);
4537 * Here with bp and block set to the leftmost leaf node in the tree.
4539 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4542 * Loop over all leaf nodes. Copy information to the extent records.
4545 xfs_bmbt_rec_t *frp, *trp;
4546 xfs_fsblock_t nextbno;
4547 xfs_extnum_t num_recs;
4551 num_recs = be16_to_cpu(block->bb_numrecs);
4552 if (unlikely(i + num_recs > room)) {
4553 ASSERT(i + num_recs <= room);
4554 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4555 "corrupt dinode %Lu, (btree extents).",
4556 (unsigned long long) ip->i_ino);
4557 XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4562 XFS_WANT_CORRUPTED_GOTO(
4563 XFS_BMAP_SANITY_CHECK(mp, block, 0),
4566 * Read-ahead the next leaf block, if any.
4568 nextbno = be64_to_cpu(block->bb_rightsib);
4569 if (nextbno != NULLFSBLOCK)
4570 xfs_btree_reada_bufl(mp, nextbno, 1);
4572 * Copy records into the extent records.
4574 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4575 block, 1, mp->m_bmap_dmxr[0]);
4577 for (j = 0; j < num_recs; j++, i++, frp++) {
4578 trp = xfs_iext_get_ext(ifp, i);
4579 trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4580 trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4582 if (exntf == XFS_EXTFMT_NOSTATE) {
4584 * Check all attribute bmap btree records and
4585 * any "older" data bmap btree records for a
4586 * set bit in the "extent flag" position.
4588 if (unlikely(xfs_check_nostate_extents(ifp,
4589 start, num_recs))) {
4590 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4596 xfs_trans_brelse(tp, bp);
4599 * If we've reached the end, stop.
4601 if (bno == NULLFSBLOCK)
4603 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4604 XFS_BMAP_BTREE_REF)))
4606 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4608 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4609 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4610 xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4613 xfs_trans_brelse(tp, bp);
4614 return XFS_ERROR(EFSCORRUPTED);
4617 #ifdef XFS_BMAP_TRACE
4619 * Add bmap trace insert entries for all the contents of the extent records.
4622 xfs_bmap_trace_exlist(
4623 char *fname, /* function name */
4624 xfs_inode_t *ip, /* incore inode pointer */
4625 xfs_extnum_t cnt, /* count of entries in the list */
4626 int whichfork) /* data or attr fork */
4628 xfs_bmbt_rec_t *ep; /* current extent record */
4629 xfs_extnum_t idx; /* extent record index */
4630 xfs_ifork_t *ifp; /* inode fork pointer */
4631 xfs_bmbt_irec_t s; /* file extent record */
4633 ifp = XFS_IFORK_PTR(ip, whichfork);
4634 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4635 for (idx = 0; idx < cnt; idx++) {
4636 ep = xfs_iext_get_ext(ifp, idx);
4637 xfs_bmbt_get_all(ep, &s);
4638 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4646 * Validate that the bmbt_irecs being returned from bmapi are valid
4647 * given the callers original parameters. Specifically check the
4648 * ranges of the returned irecs to ensure that they only extent beyond
4649 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4652 xfs_bmap_validate_ret(
4656 xfs_bmbt_irec_t *mval,
4660 int i; /* index to map values */
4662 ASSERT(ret_nmap <= nmap);
4664 for (i = 0; i < ret_nmap; i++) {
4665 ASSERT(mval[i].br_blockcount > 0);
4666 if (!(flags & XFS_BMAPI_ENTIRE)) {
4667 ASSERT(mval[i].br_startoff >= bno);
4668 ASSERT(mval[i].br_blockcount <= len);
4669 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4672 ASSERT(mval[i].br_startoff < bno + len);
4673 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4677 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4678 mval[i].br_startoff);
4679 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4680 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4681 mval[i].br_startblock != HOLESTARTBLOCK);
4682 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4683 mval[i].br_state == XFS_EXT_UNWRITTEN);
4690 * Map file blocks to filesystem blocks.
4691 * File range is given by the bno/len pair.
4692 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4693 * into a hole or past eof.
4694 * Only allocates blocks from a single allocation group,
4695 * to avoid locking problems.
4696 * The returned value in "firstblock" from the first call in a transaction
4697 * must be remembered and presented to subsequent calls in "firstblock".
4698 * An upper bound for the number of blocks to be allocated is supplied to
4699 * the first call in "total"; if no allocation group has that many free
4700 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4704 xfs_trans_t *tp, /* transaction pointer */
4705 xfs_inode_t *ip, /* incore inode */
4706 xfs_fileoff_t bno, /* starting file offs. mapped */
4707 xfs_filblks_t len, /* length to map in file */
4708 int flags, /* XFS_BMAPI_... */
4709 xfs_fsblock_t *firstblock, /* first allocated block
4710 controls a.g. for allocs */
4711 xfs_extlen_t total, /* total blocks needed */
4712 xfs_bmbt_irec_t *mval, /* output: map values */
4713 int *nmap, /* i/o: mval size/count */
4714 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4715 xfs_extdelta_t *delta) /* o: change made to incore extents */
4717 xfs_fsblock_t abno; /* allocated block number */
4718 xfs_extlen_t alen; /* allocated extent length */
4719 xfs_fileoff_t aoff; /* allocated file offset */
4720 xfs_bmalloca_t bma; /* args for xfs_bmap_alloc */
4721 xfs_btree_cur_t *cur; /* bmap btree cursor */
4722 xfs_fileoff_t end; /* end of mapped file region */
4723 int eof; /* we've hit the end of extents */
4724 xfs_bmbt_rec_t *ep; /* extent record pointer */
4725 int error; /* error return */
4726 xfs_bmbt_irec_t got; /* current file extent record */
4727 xfs_ifork_t *ifp; /* inode fork pointer */
4728 xfs_extlen_t indlen; /* indirect blocks length */
4729 xfs_extnum_t lastx; /* last useful extent number */
4730 int logflags; /* flags for transaction logging */
4731 xfs_extlen_t minleft; /* min blocks left after allocation */
4732 xfs_extlen_t minlen; /* min allocation size */
4733 xfs_mount_t *mp; /* xfs mount structure */
4734 int n; /* current extent index */
4735 int nallocs; /* number of extents alloc\'d */
4736 xfs_extnum_t nextents; /* number of extents in file */
4737 xfs_fileoff_t obno; /* old block number (offset) */
4738 xfs_bmbt_irec_t prev; /* previous file extent record */
4739 int tmp_logflags; /* temp flags holder */
4740 int whichfork; /* data or attr fork */
4741 char inhole; /* current location is hole in file */
4742 char wasdelay; /* old extent was delayed */
4743 char wr; /* this is a write request */
4744 char rt; /* this is a realtime file */
4746 xfs_fileoff_t orig_bno; /* original block number value */
4747 int orig_flags; /* original flags arg value */
4748 xfs_filblks_t orig_len; /* original value of len arg */
4749 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4750 int orig_nmap; /* original value of *nmap */
4759 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4760 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4761 XFS_ATTR_FORK : XFS_DATA_FORK;
4763 if (unlikely(XFS_TEST_ERROR(
4764 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4765 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4766 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4767 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4768 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4769 return XFS_ERROR(EFSCORRUPTED);
4771 if (XFS_FORCED_SHUTDOWN(mp))
4772 return XFS_ERROR(EIO);
4773 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4774 ifp = XFS_IFORK_PTR(ip, whichfork);
4775 ASSERT(ifp->if_ext_max ==
4776 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4777 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4778 XFS_STATS_INC(xs_blk_mapw);
4780 XFS_STATS_INC(xs_blk_mapr);
4782 * IGSTATE flag is used to combine extents which
4783 * differ only due to the state of the extents.
4784 * This technique is used from xfs_getbmap()
4785 * when the caller does not wish to see the
4786 * separation (which is the default).
4788 * This technique is also used when writing a
4789 * buffer which has been partially written,
4790 * (usually by being flushed during a chunkread),
4791 * to ensure one write takes place. This also
4792 * prevents a change in the xfs inode extents at
4793 * this time, intentionally. This change occurs
4794 * on completion of the write operation, in
4795 * xfs_strat_comp(), where the xfs_bmapi() call
4796 * is transactioned, and the extents combined.
4798 if ((flags & XFS_BMAPI_IGSTATE) && wr) /* if writing unwritten space */
4799 wr = 0; /* no allocations are allowed */
4800 ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4804 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4806 if ((error = xfs_bmap_local_to_extents(tp, ip,
4807 firstblock, total, &logflags, whichfork)))
4810 if (wr && *firstblock == NULLFSBLOCK) {
4811 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4812 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4817 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4818 (error = xfs_iread_extents(tp, ip, whichfork)))
4820 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4822 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4828 delta->xed_startoff = NULLFILEOFF;
4829 delta->xed_blockcount = 0;
4831 while (bno < end && n < *nmap) {
4833 * Reading past eof, act as though there's a hole
4837 got.br_startoff = end;
4838 inhole = eof || got.br_startoff > bno;
4839 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4840 ISNULLSTARTBLOCK(got.br_startblock);
4842 * First, deal with the hole before the allocated space
4843 * that we found, if any.
4845 if (wr && (inhole || wasdelay)) {
4847 * For the wasdelay case, we could also just
4848 * allocate the stuff asked for in this bmap call
4849 * but that wouldn't be as good.
4851 if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4852 alen = (xfs_extlen_t)got.br_blockcount;
4853 aoff = got.br_startoff;
4854 if (lastx != NULLEXTNUM && lastx) {
4855 ep = xfs_iext_get_ext(ifp, lastx - 1);
4856 xfs_bmbt_get_all(ep, &prev);
4858 } else if (wasdelay) {
4859 alen = (xfs_extlen_t)
4860 XFS_FILBLKS_MIN(len,
4862 got.br_blockcount) - bno);
4865 alen = (xfs_extlen_t)
4866 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4868 alen = (xfs_extlen_t)
4869 XFS_FILBLKS_MIN(alen,
4870 got.br_startoff - bno);
4873 minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4874 if (flags & XFS_BMAPI_DELAY) {
4877 /* Figure out the extent size, adjust alen */
4879 if (!(extsz = ip->i_d.di_extsize))
4880 extsz = mp->m_sb.sb_rextsize;
4882 extsz = ip->i_d.di_extsize;
4885 error = xfs_bmap_extsize_align(mp,
4888 flags&XFS_BMAPI_DELAY,
4889 flags&XFS_BMAPI_CONVERT,
4895 extsz = alen / mp->m_sb.sb_rextsize;
4898 * Make a transaction-less quota reservation for
4899 * delayed allocation blocks. This number gets
4900 * adjusted later. We return if we haven't
4901 * allocated blocks already inside this loop.
4903 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4904 mp, NULL, ip, (long)alen, 0,
4905 rt ? XFS_QMOPT_RES_RTBLKS :
4906 XFS_QMOPT_RES_REGBLKS))) {
4909 ASSERT(cur == NULL);
4916 * Split changing sb for alen and indlen since
4917 * they could be coming from different places.
4919 indlen = (xfs_extlen_t)
4920 xfs_bmap_worst_indlen(ip, alen);
4924 error = xfs_mod_incore_sb(mp,
4927 XFS_BMAPI_RSVBLOCKS));
4929 error = xfs_mod_incore_sb(mp,
4932 XFS_BMAPI_RSVBLOCKS));
4935 error = xfs_mod_incore_sb(mp,
4938 XFS_BMAPI_RSVBLOCKS));
4940 xfs_mod_incore_sb(mp,
4943 XFS_BMAPI_RSVBLOCKS));
4945 xfs_mod_incore_sb(mp,
4948 XFS_BMAPI_RSVBLOCKS));
4952 if (XFS_IS_QUOTA_ON(mp))
4953 /* unreserve the blocks now */
4955 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4958 XFS_QMOPT_RES_RTBLKS :
4959 XFS_QMOPT_RES_REGBLKS);
4963 ip->i_delayed_blks += alen;
4964 abno = NULLSTARTBLOCK(indlen);
4967 * If first time, allocate and fill in
4968 * once-only bma fields.
4970 if (bma.ip == NULL) {
4978 /* Indicate if this is the first user data
4979 * in the file, or just any user data.
4981 if (!(flags & XFS_BMAPI_METADATA)) {
4982 bma.userdata = (aoff == 0) ?
4983 XFS_ALLOC_INITIAL_USER_DATA :
4987 * Fill in changeable bma fields.
4990 bma.firstblock = *firstblock;
4993 bma.conv = (flags & XFS_BMAPI_CONVERT);
4994 bma.wasdel = wasdelay;
4995 bma.minlen = minlen;
4996 bma.low = flist->xbf_low;
4997 bma.minleft = minleft;
4999 * Only want to do the alignment at the
5000 * eof if it is userdata and allocation length
5001 * is larger than a stripe unit.
5003 if (mp->m_dalign && alen >= mp->m_dalign &&
5004 (!(flags & XFS_BMAPI_METADATA)) &&
5005 (whichfork == XFS_DATA_FORK)) {
5006 if ((error = xfs_bmap_isaeof(ip, aoff,
5007 whichfork, &bma.aeof)))
5014 if ((error = xfs_bmap_alloc(&bma)))
5017 * Copy out result fields.
5020 if ((flist->xbf_low = bma.low))
5024 ASSERT(*firstblock == NULLFSBLOCK ||
5025 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5026 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5028 XFS_FSB_TO_AGNO(mp, *firstblock) <
5029 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5030 *firstblock = bma.firstblock;
5032 cur->bc_private.b.firstblock =
5034 if (abno == NULLFSBLOCK)
5036 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5037 cur = xfs_btree_init_cursor(mp,
5038 tp, NULL, 0, XFS_BTNUM_BMAP,
5040 cur->bc_private.b.firstblock =
5042 cur->bc_private.b.flist = flist;
5045 * Bump the number of extents we've allocated
5051 cur->bc_private.b.flags =
5052 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5053 got.br_startoff = aoff;
5054 got.br_startblock = abno;
5055 got.br_blockcount = alen;
5056 got.br_state = XFS_EXT_NORM; /* assume normal */
5058 * Determine state of extent, and the filesystem.
5059 * A wasdelay extent has been initialized, so
5060 * shouldn't be flagged as unwritten.
5062 if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5063 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5064 got.br_state = XFS_EXT_UNWRITTEN;
5066 error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5067 firstblock, flist, &tmp_logflags, delta,
5068 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5069 logflags |= tmp_logflags;
5072 lastx = ifp->if_lastex;
5073 ep = xfs_iext_get_ext(ifp, lastx);
5074 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5075 xfs_bmbt_get_all(ep, &got);
5076 ASSERT(got.br_startoff <= aoff);
5077 ASSERT(got.br_startoff + got.br_blockcount >=
5080 if (flags & XFS_BMAPI_DELAY) {
5081 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5082 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5084 ASSERT(got.br_state == XFS_EXT_NORM ||
5085 got.br_state == XFS_EXT_UNWRITTEN);
5088 * Fall down into the found allocated space case.
5090 } else if (inhole) {
5092 * Reading in a hole.
5094 mval->br_startoff = bno;
5095 mval->br_startblock = HOLESTARTBLOCK;
5096 mval->br_blockcount =
5097 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5098 mval->br_state = XFS_EXT_NORM;
5099 bno += mval->br_blockcount;
5100 len -= mval->br_blockcount;
5106 * Then deal with the allocated space we found.
5109 if (!(flags & XFS_BMAPI_ENTIRE) &&
5110 (got.br_startoff + got.br_blockcount > obno)) {
5113 ASSERT((bno >= obno) || (n == 0));
5115 mval->br_startoff = bno;
5116 if (ISNULLSTARTBLOCK(got.br_startblock)) {
5117 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5118 mval->br_startblock = DELAYSTARTBLOCK;
5120 mval->br_startblock =
5122 (bno - got.br_startoff);
5124 * Return the minimum of what we got and what we
5125 * asked for for the length. We can use the len
5126 * variable here because it is modified below
5127 * and we could have been there before coming
5128 * here if the first part of the allocation
5129 * didn't overlap what was asked for.
5131 mval->br_blockcount =
5132 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5133 (bno - got.br_startoff));
5134 mval->br_state = got.br_state;
5135 ASSERT(mval->br_blockcount <= len);
5138 if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5139 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5140 mval->br_startblock = DELAYSTARTBLOCK;
5145 * Check if writing previously allocated but
5146 * unwritten extents.
5148 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5149 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5151 * Modify (by adding) the state flag, if writing.
5153 ASSERT(mval->br_blockcount <= len);
5154 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5155 cur = xfs_btree_init_cursor(mp,
5156 tp, NULL, 0, XFS_BTNUM_BMAP,
5158 cur->bc_private.b.firstblock =
5160 cur->bc_private.b.flist = flist;
5162 mval->br_state = XFS_EXT_NORM;
5163 error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5164 firstblock, flist, &tmp_logflags, delta,
5165 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5166 logflags |= tmp_logflags;
5169 lastx = ifp->if_lastex;
5170 ep = xfs_iext_get_ext(ifp, lastx);
5171 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5172 xfs_bmbt_get_all(ep, &got);
5174 * We may have combined previously unwritten
5175 * space with written space, so generate
5178 if (mval->br_blockcount < len)
5182 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5183 ((mval->br_startoff + mval->br_blockcount) <= end));
5184 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5185 (mval->br_blockcount <= len) ||
5186 (mval->br_startoff < obno));
5187 bno = mval->br_startoff + mval->br_blockcount;
5189 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5190 ASSERT(mval->br_startblock == mval[-1].br_startblock);
5191 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5192 ASSERT(mval->br_state == mval[-1].br_state);
5193 mval[-1].br_blockcount = mval->br_blockcount;
5194 mval[-1].br_state = mval->br_state;
5195 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5196 mval[-1].br_startblock != DELAYSTARTBLOCK &&
5197 mval[-1].br_startblock != HOLESTARTBLOCK &&
5198 mval->br_startblock ==
5199 mval[-1].br_startblock + mval[-1].br_blockcount &&
5200 ((flags & XFS_BMAPI_IGSTATE) ||
5201 mval[-1].br_state == mval->br_state)) {
5202 ASSERT(mval->br_startoff ==
5203 mval[-1].br_startoff + mval[-1].br_blockcount);
5204 mval[-1].br_blockcount += mval->br_blockcount;
5206 mval->br_startblock == DELAYSTARTBLOCK &&
5207 mval[-1].br_startblock == DELAYSTARTBLOCK &&
5208 mval->br_startoff ==
5209 mval[-1].br_startoff + mval[-1].br_blockcount) {
5210 mval[-1].br_blockcount += mval->br_blockcount;
5211 mval[-1].br_state = mval->br_state;
5212 } else if (!((n == 0) &&
5213 ((mval->br_startoff + mval->br_blockcount) <=
5219 * If we're done, stop now. Stop when we've allocated
5220 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5221 * the transaction may get too big.
5223 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5226 * Else go on to the next record.
5228 ep = xfs_iext_get_ext(ifp, ++lastx);
5229 if (lastx >= nextents) {
5233 xfs_bmbt_get_all(ep, &got);
5235 ifp->if_lastex = lastx;
5238 * Transform from btree to extents, give it cur.
5240 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5241 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5243 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5244 &tmp_logflags, whichfork);
5245 logflags |= tmp_logflags;
5249 ASSERT(ifp->if_ext_max ==
5250 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5251 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5252 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5254 if (delta && delta->xed_startoff != NULLFILEOFF) {
5255 /* A change was actually made.
5256 * Note that delta->xed_blockount is an offset at this
5257 * point and needs to be converted to a block count.
5259 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5260 delta->xed_blockcount -= delta->xed_startoff;
5264 * Log everything. Do this after conversion, there's no point in
5265 * logging the extent records if we've converted to btree format.
5267 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5268 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5269 logflags &= ~XFS_ILOG_FEXT(whichfork);
5270 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5271 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5272 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5274 * Log whatever the flags say, even if error. Otherwise we might miss
5275 * detecting a case where the data is changed, there's an error,
5276 * and it's not logged so we don't shutdown when we should.
5280 xfs_trans_log_inode(tp, ip, logflags);
5284 ASSERT(*firstblock == NULLFSBLOCK ||
5285 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5287 cur->bc_private.b.firstblock) ||
5289 XFS_FSB_TO_AGNO(mp, *firstblock) <
5291 cur->bc_private.b.firstblock)));
5292 *firstblock = cur->bc_private.b.firstblock;
5294 xfs_btree_del_cursor(cur,
5295 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5298 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5304 * Map file blocks to filesystem blocks, simple version.
5305 * One block (extent) only, read-only.
5306 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5307 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5308 * was set and all the others were clear.
5312 xfs_trans_t *tp, /* transaction pointer */
5313 xfs_inode_t *ip, /* incore inode */
5314 int whichfork, /* data or attr fork */
5315 xfs_fsblock_t *fsb, /* output: mapped block */
5316 xfs_fileoff_t bno) /* starting file offs. mapped */
5318 int eof; /* we've hit the end of extents */
5319 int error; /* error return */
5320 xfs_bmbt_irec_t got; /* current file extent record */
5321 xfs_ifork_t *ifp; /* inode fork pointer */
5322 xfs_extnum_t lastx; /* last useful extent number */
5323 xfs_bmbt_irec_t prev; /* previous file extent record */
5325 ifp = XFS_IFORK_PTR(ip, whichfork);
5327 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5328 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5329 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5331 return XFS_ERROR(EFSCORRUPTED);
5333 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5334 return XFS_ERROR(EIO);
5335 XFS_STATS_INC(xs_blk_mapr);
5336 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5337 (error = xfs_iread_extents(tp, ip, whichfork)))
5339 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5342 * Reading past eof, act as though there's a hole
5345 if (eof || got.br_startoff > bno) {
5349 ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5350 ASSERT(bno < got.br_startoff + got.br_blockcount);
5351 *fsb = got.br_startblock + (bno - got.br_startoff);
5352 ifp->if_lastex = lastx;
5357 * Unmap (remove) blocks from a file.
5358 * If nexts is nonzero then the number of extents to remove is limited to
5359 * that value. If not all extents in the block range can be removed then
5364 xfs_trans_t *tp, /* transaction pointer */
5365 struct xfs_inode *ip, /* incore inode */
5366 xfs_fileoff_t bno, /* starting offset to unmap */
5367 xfs_filblks_t len, /* length to unmap in file */
5368 int flags, /* misc flags */
5369 xfs_extnum_t nexts, /* number of extents max */
5370 xfs_fsblock_t *firstblock, /* first allocated block
5371 controls a.g. for allocs */
5372 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5373 xfs_extdelta_t *delta, /* o: change made to incore
5375 int *done) /* set if not done yet */
5377 xfs_btree_cur_t *cur; /* bmap btree cursor */
5378 xfs_bmbt_irec_t del; /* extent being deleted */
5379 int eof; /* is deleting at eof */
5380 xfs_bmbt_rec_t *ep; /* extent record pointer */
5381 int error; /* error return value */
5382 xfs_extnum_t extno; /* extent number in list */
5383 xfs_bmbt_irec_t got; /* current extent record */
5384 xfs_ifork_t *ifp; /* inode fork pointer */
5385 int isrt; /* freeing in rt area */
5386 xfs_extnum_t lastx; /* last extent index used */
5387 int logflags; /* transaction logging flags */
5388 xfs_extlen_t mod; /* rt extent offset */
5389 xfs_mount_t *mp; /* mount structure */
5390 xfs_extnum_t nextents; /* number of file extents */
5391 xfs_bmbt_irec_t prev; /* previous extent record */
5392 xfs_fileoff_t start; /* first file offset deleted */
5393 int tmp_logflags; /* partial logging flags */
5394 int wasdel; /* was a delayed alloc extent */
5395 int whichfork; /* data or attribute fork */
5396 int rsvd; /* OK to allocate reserved blocks */
5399 xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5400 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5401 XFS_ATTR_FORK : XFS_DATA_FORK;
5402 ifp = XFS_IFORK_PTR(ip, whichfork);
5404 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5405 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5406 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5408 return XFS_ERROR(EFSCORRUPTED);
5411 if (XFS_FORCED_SHUTDOWN(mp))
5412 return XFS_ERROR(EIO);
5413 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5416 ASSERT(ifp->if_ext_max ==
5417 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5418 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5419 (error = xfs_iread_extents(tp, ip, whichfork)))
5421 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5422 if (nextents == 0) {
5426 XFS_STATS_INC(xs_blk_unmap);
5427 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5429 bno = start + len - 1;
5430 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5433 delta->xed_startoff = NULLFILEOFF;
5434 delta->xed_blockcount = 0;
5437 * Check to see if the given block number is past the end of the
5438 * file, back up to the last block if so...
5441 ep = xfs_iext_get_ext(ifp, --lastx);
5442 xfs_bmbt_get_all(ep, &got);
5443 bno = got.br_startoff + got.br_blockcount - 1;
5446 if (ifp->if_flags & XFS_IFBROOT) {
5447 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5448 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5450 cur->bc_private.b.firstblock = *firstblock;
5451 cur->bc_private.b.flist = flist;
5452 cur->bc_private.b.flags = 0;
5456 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5457 (nexts == 0 || extno < nexts)) {
5459 * Is the found extent after a hole in which bno lives?
5460 * Just back up to the previous extent, if so.
5462 if (got.br_startoff > bno) {
5465 ep = xfs_iext_get_ext(ifp, lastx);
5466 xfs_bmbt_get_all(ep, &got);
5469 * Is the last block of this extent before the range
5470 * we're supposed to delete? If so, we're done.
5472 bno = XFS_FILEOFF_MIN(bno,
5473 got.br_startoff + got.br_blockcount - 1);
5477 * Then deal with the (possibly delayed) allocated space
5482 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5483 if (got.br_startoff < start) {
5484 del.br_startoff = start;
5485 del.br_blockcount -= start - got.br_startoff;
5487 del.br_startblock += start - got.br_startoff;
5489 if (del.br_startoff + del.br_blockcount > bno + 1)
5490 del.br_blockcount = bno + 1 - del.br_startoff;
5491 sum = del.br_startblock + del.br_blockcount;
5493 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5495 * Realtime extent not lined up at the end.
5496 * The extent could have been split into written
5497 * and unwritten pieces, or we could just be
5498 * unmapping part of it. But we can't really
5499 * get rid of part of a realtime extent.
5501 if (del.br_state == XFS_EXT_UNWRITTEN ||
5502 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5504 * This piece is unwritten, or we're not
5505 * using unwritten extents. Skip over it.
5508 bno -= mod > del.br_blockcount ?
5509 del.br_blockcount : mod;
5510 if (bno < got.br_startoff) {
5512 xfs_bmbt_get_all(xfs_iext_get_ext(
5518 * It's written, turn it unwritten.
5519 * This is better than zeroing it.
5521 ASSERT(del.br_state == XFS_EXT_NORM);
5522 ASSERT(xfs_trans_get_block_res(tp) > 0);
5524 * If this spans a realtime extent boundary,
5525 * chop it back to the start of the one we end at.
5527 if (del.br_blockcount > mod) {
5528 del.br_startoff += del.br_blockcount - mod;
5529 del.br_startblock += del.br_blockcount - mod;
5530 del.br_blockcount = mod;
5532 del.br_state = XFS_EXT_UNWRITTEN;
5533 error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5534 firstblock, flist, &logflags, delta,
5540 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5542 * Realtime extent is lined up at the end but not
5543 * at the front. We'll get rid of full extents if
5546 mod = mp->m_sb.sb_rextsize - mod;
5547 if (del.br_blockcount > mod) {
5548 del.br_blockcount -= mod;
5549 del.br_startoff += mod;
5550 del.br_startblock += mod;
5551 } else if ((del.br_startoff == start &&
5552 (del.br_state == XFS_EXT_UNWRITTEN ||
5553 xfs_trans_get_block_res(tp) == 0)) ||
5554 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5556 * Can't make it unwritten. There isn't
5557 * a full extent here so just skip it.
5559 ASSERT(bno >= del.br_blockcount);
5560 bno -= del.br_blockcount;
5561 if (bno < got.br_startoff) {
5563 xfs_bmbt_get_all(--ep, &got);
5566 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5568 * This one is already unwritten.
5569 * It must have a written left neighbor.
5570 * Unwrite the killed part of that one and
5574 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5576 ASSERT(prev.br_state == XFS_EXT_NORM);
5577 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5578 ASSERT(del.br_startblock ==
5579 prev.br_startblock + prev.br_blockcount);
5580 if (prev.br_startoff < start) {
5581 mod = start - prev.br_startoff;
5582 prev.br_blockcount -= mod;
5583 prev.br_startblock += mod;
5584 prev.br_startoff = start;
5586 prev.br_state = XFS_EXT_UNWRITTEN;
5587 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5588 &prev, firstblock, flist, &logflags,
5589 delta, XFS_DATA_FORK, 0);
5594 ASSERT(del.br_state == XFS_EXT_NORM);
5595 del.br_state = XFS_EXT_UNWRITTEN;
5596 error = xfs_bmap_add_extent(ip, lastx, &cur,
5597 &del, firstblock, flist, &logflags,
5598 delta, XFS_DATA_FORK, 0);
5605 ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5606 /* Update realtime/data freespace, unreserve quota */
5608 xfs_filblks_t rtexts;
5610 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5611 do_div(rtexts, mp->m_sb.sb_rextsize);
5612 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5614 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5615 NULL, ip, -((long)del.br_blockcount), 0,
5616 XFS_QMOPT_RES_RTBLKS);
5618 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5619 (int)del.br_blockcount, rsvd);
5620 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5621 NULL, ip, -((long)del.br_blockcount), 0,
5622 XFS_QMOPT_RES_REGBLKS);
5624 ip->i_delayed_blks -= del.br_blockcount;
5626 cur->bc_private.b.flags |=
5627 XFS_BTCUR_BPRV_WASDEL;
5629 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5631 * If it's the case where the directory code is running
5632 * with no block reservation, and the deleted block is in
5633 * the middle of its extent, and the resulting insert
5634 * of an extent would cause transformation to btree format,
5635 * then reject it. The calling code will then swap
5636 * blocks around instead.
5637 * We have to do this now, rather than waiting for the
5638 * conversion to btree format, since the transaction
5641 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5642 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5643 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5644 del.br_startoff > got.br_startoff &&
5645 del.br_startoff + del.br_blockcount <
5646 got.br_startoff + got.br_blockcount) {
5647 error = XFS_ERROR(ENOSPC);
5650 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5651 &tmp_logflags, delta, whichfork, rsvd);
5652 logflags |= tmp_logflags;
5655 bno = del.br_startoff - 1;
5657 lastx = ifp->if_lastex;
5659 * If not done go on to the next (previous) record.
5660 * Reset ep in case the extents array was re-alloced.
5662 ep = xfs_iext_get_ext(ifp, lastx);
5663 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5664 if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5665 xfs_bmbt_get_startoff(ep) > bno) {
5667 ep = xfs_iext_get_ext(ifp, lastx);
5670 xfs_bmbt_get_all(ep, &got);
5674 ifp->if_lastex = lastx;
5675 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5676 ASSERT(ifp->if_ext_max ==
5677 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5679 * Convert to a btree if necessary.
5681 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5682 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5683 ASSERT(cur == NULL);
5684 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5685 &cur, 0, &tmp_logflags, whichfork);
5686 logflags |= tmp_logflags;
5691 * transform from btree to extents, give it cur
5693 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5694 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5695 ASSERT(cur != NULL);
5696 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5698 logflags |= tmp_logflags;
5703 * transform from extents to local?
5705 ASSERT(ifp->if_ext_max ==
5706 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5708 if (delta && delta->xed_startoff != NULLFILEOFF) {
5709 /* A change was actually made.
5710 * Note that delta->xed_blockount is an offset at this
5711 * point and needs to be converted to a block count.
5713 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5714 delta->xed_blockcount -= delta->xed_startoff;
5718 * Log everything. Do this after conversion, there's no point in
5719 * logging the extent records if we've converted to btree format.
5721 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5722 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5723 logflags &= ~XFS_ILOG_FEXT(whichfork);
5724 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5725 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5726 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5728 * Log inode even in the error case, if the transaction
5729 * is dirty we'll need to shut down the filesystem.
5732 xfs_trans_log_inode(tp, ip, logflags);
5735 *firstblock = cur->bc_private.b.firstblock;
5736 cur->bc_private.b.allocated = 0;
5738 xfs_btree_del_cursor(cur,
5739 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5745 * Fcntl interface to xfs_bmapi.
5747 int /* error code */
5749 bhv_desc_t *bdp, /* XFS behavior descriptor*/
5750 struct getbmap *bmv, /* user bmap structure */
5751 void __user *ap, /* pointer to user's array */
5752 int interface) /* interface flags */
5754 __int64_t bmvend; /* last block requested */
5755 int error; /* return value */
5756 __int64_t fixlen; /* length for -1 case */
5757 int i; /* extent number */
5758 xfs_inode_t *ip; /* xfs incore inode pointer */
5759 vnode_t *vp; /* corresponding vnode */
5760 int lock; /* lock state */
5761 xfs_bmbt_irec_t *map; /* buffer for user's data */
5762 xfs_mount_t *mp; /* file system mount point */
5763 int nex; /* # of user extents can do */
5764 int nexleft; /* # of user extents left */
5765 int subnex; /* # of bmapi's can do */
5766 int nmap; /* number of map entries */
5767 struct getbmap out; /* output structure */
5768 int whichfork; /* data or attr fork */
5769 int prealloced; /* this is a file with
5770 * preallocated data space */
5771 int sh_unwritten; /* true, if unwritten */
5772 /* extents listed separately */
5773 int bmapi_flags; /* flags for xfs_bmapi */
5774 __int32_t oflags; /* getbmapx bmv_oflags field */
5776 vp = BHV_TO_VNODE(bdp);
5777 ip = XFS_BHVTOI(bdp);
5780 whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5781 sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5783 /* If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5784 * generate a DMAPI read event. Otherwise, if the DM_EVENT_READ
5785 * bit is set for the file, generate a read event in order
5786 * that the DMAPI application may do its thing before we return
5787 * the extents. Usually this means restoring user file data to
5788 * regions of the file that look like holes.
5790 * The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5791 * BMV_IF_NO_DMAPI_READ so that read events are generated.
5792 * If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5793 * could misinterpret holes in a DMAPI file as true holes,
5794 * when in fact they may represent offline user data.
5796 if ( (interface & BMV_IF_NO_DMAPI_READ) == 0
5797 && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5798 && whichfork == XFS_DATA_FORK) {
5800 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5802 return XFS_ERROR(error);
5805 if (whichfork == XFS_ATTR_FORK) {
5806 if (XFS_IFORK_Q(ip)) {
5807 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5808 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5809 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5810 return XFS_ERROR(EINVAL);
5811 } else if (unlikely(
5812 ip->i_d.di_aformat != 0 &&
5813 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5814 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5816 return XFS_ERROR(EFSCORRUPTED);
5818 } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5819 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5820 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5821 return XFS_ERROR(EINVAL);
5822 if (whichfork == XFS_DATA_FORK) {
5823 if ((ip->i_d.di_extsize && (ip->i_d.di_flags &
5824 (XFS_DIFLAG_REALTIME|XFS_DIFLAG_EXTSIZE))) ||
5825 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5827 fixlen = XFS_MAXIOFFSET(mp);
5830 fixlen = ip->i_d.di_size;
5837 if (bmv->bmv_length == -1) {
5838 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5839 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5841 } else if (bmv->bmv_length < 0)
5842 return XFS_ERROR(EINVAL);
5843 if (bmv->bmv_length == 0) {
5844 bmv->bmv_entries = 0;
5847 nex = bmv->bmv_count - 1;
5849 return XFS_ERROR(EINVAL);
5850 bmvend = bmv->bmv_offset + bmv->bmv_length;
5852 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5854 if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5855 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5856 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5859 ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5861 lock = xfs_ilock_map_shared(ip);
5864 * Don't let nex be bigger than the number of extents
5865 * we can have assuming alternating holes and real extents.
5867 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5868 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5870 bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5871 ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5874 * Allocate enough space to handle "subnex" maps at a time.
5877 map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5879 bmv->bmv_entries = 0;
5881 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5883 goto unlock_and_return;
5889 nmap = (nexleft > subnex) ? subnex : nexleft;
5890 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5891 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5892 bmapi_flags, NULL, 0, map, &nmap,
5895 goto unlock_and_return;
5896 ASSERT(nmap <= subnex);
5898 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5900 oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5901 BMV_OF_PREALLOC : 0;
5902 out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5903 out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5904 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5905 if (map[i].br_startblock == HOLESTARTBLOCK &&
5906 ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5907 whichfork == XFS_ATTR_FORK )) {
5909 * came to hole at end of file or the end of
5912 goto unlock_and_return;
5915 (map[i].br_startblock == HOLESTARTBLOCK) ?
5917 XFS_FSB_TO_DB(ip, map[i].br_startblock);
5919 /* return either getbmap/getbmapx structure. */
5920 if (interface & BMV_IF_EXTENDED) {
5921 struct getbmapx outx;
5923 GETBMAP_CONVERT(out,outx);
5924 outx.bmv_oflags = oflags;
5925 outx.bmv_unused1 = outx.bmv_unused2 = 0;
5926 if (copy_to_user(ap, &outx,
5928 error = XFS_ERROR(EFAULT);
5929 goto unlock_and_return;
5932 if (copy_to_user(ap, &out,
5934 error = XFS_ERROR(EFAULT);
5935 goto unlock_and_return;
5939 out.bmv_offset + out.bmv_length;
5940 bmv->bmv_length = MAX((__int64_t)0,
5941 (__int64_t)(bmvend - bmv->bmv_offset));
5943 ap = (interface & BMV_IF_EXTENDED) ?
5945 ((struct getbmapx __user *)ap + 1) :
5947 ((struct getbmap __user *)ap + 1);
5950 } while (nmap && nexleft && bmv->bmv_length);
5953 xfs_iunlock_map_shared(ip, lock);
5954 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5956 kmem_free(map, subnex * sizeof(*map));
5962 * Check the last inode extent to determine whether this allocation will result
5963 * in blocks being allocated at the end of the file. When we allocate new data
5964 * blocks at the end of the file which do not start at the previous data block,
5965 * we will try to align the new blocks at stripe unit boundaries.
5967 STATIC int /* error */
5969 xfs_inode_t *ip, /* incore inode pointer */
5970 xfs_fileoff_t off, /* file offset in fsblocks */
5971 int whichfork, /* data or attribute fork */
5972 char *aeof) /* return value */
5974 int error; /* error return value */
5975 xfs_ifork_t *ifp; /* inode fork pointer */
5976 xfs_bmbt_rec_t *lastrec; /* extent record pointer */
5977 xfs_extnum_t nextents; /* number of file extents */
5978 xfs_bmbt_irec_t s; /* expanded extent record */
5980 ASSERT(whichfork == XFS_DATA_FORK);
5981 ifp = XFS_IFORK_PTR(ip, whichfork);
5982 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5983 (error = xfs_iread_extents(NULL, ip, whichfork)))
5985 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5986 if (nextents == 0) {
5991 * Go to the last extent
5993 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5994 xfs_bmbt_get_all(lastrec, &s);
5996 * Check we are allocating in the last extent (for delayed allocations)
5997 * or past the last extent for non-delayed allocations.
5999 *aeof = (off >= s.br_startoff &&
6000 off < s.br_startoff + s.br_blockcount &&
6001 ISNULLSTARTBLOCK(s.br_startblock)) ||
6002 off >= s.br_startoff + s.br_blockcount;
6007 * Check if the endoff is outside the last extent. If so the caller will grow
6008 * the allocation to a stripe unit boundary.
6012 xfs_inode_t *ip, /* incore inode pointer */
6013 xfs_fileoff_t endoff, /* file offset in fsblocks */
6014 int whichfork, /* data or attribute fork */
6015 int *eof) /* result value */
6017 xfs_fsblock_t blockcount; /* extent block count */
6018 int error; /* error return value */
6019 xfs_ifork_t *ifp; /* inode fork pointer */
6020 xfs_bmbt_rec_t *lastrec; /* extent record pointer */
6021 xfs_extnum_t nextents; /* number of file extents */
6022 xfs_fileoff_t startoff; /* extent starting file offset */
6024 ASSERT(whichfork == XFS_DATA_FORK);
6025 ifp = XFS_IFORK_PTR(ip, whichfork);
6026 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6027 (error = xfs_iread_extents(NULL, ip, whichfork)))
6029 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6030 if (nextents == 0) {
6035 * Go to the last extent
6037 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6038 startoff = xfs_bmbt_get_startoff(lastrec);
6039 blockcount = xfs_bmbt_get_blockcount(lastrec);
6040 *eof = endoff >= startoff + blockcount;
6046 * Check that the extents list for the inode ip is in the right order.
6049 xfs_bmap_check_extents(
6050 xfs_inode_t *ip, /* incore inode pointer */
6051 int whichfork) /* data or attr fork */
6053 xfs_bmbt_rec_t *ep; /* current extent entry */
6054 xfs_extnum_t idx; /* extent record index */
6055 xfs_ifork_t *ifp; /* inode fork pointer */
6056 xfs_extnum_t nextents; /* number of extents in list */
6057 xfs_bmbt_rec_t *nextp; /* next extent entry */
6059 ifp = XFS_IFORK_PTR(ip, whichfork);
6060 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
6061 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6062 ep = xfs_iext_get_ext(ifp, 0);
6063 for (idx = 0; idx < nextents - 1; idx++) {
6064 nextp = xfs_iext_get_ext(ifp, idx + 1);
6065 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6074 xfs_btree_cur_t *cur,
6084 for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6085 bp = cur->bc_bufs[i];
6087 if (XFS_BUF_ADDR(bp) == bno)
6088 break; /* Found it */
6090 if (i == XFS_BTREE_MAXLEVELS)
6093 if (!bp) { /* Chase down all the log items to see if the bp is there */
6094 xfs_log_item_chunk_t *licp;
6098 licp = &tp->t_items;
6099 while (!bp && licp != NULL) {
6100 if (XFS_LIC_ARE_ALL_FREE(licp)) {
6101 licp = licp->lic_next;
6104 for (i = 0; i < licp->lic_unused; i++) {
6105 xfs_log_item_desc_t *lidp;
6106 xfs_log_item_t *lip;
6107 xfs_buf_log_item_t *bip;
6110 if (XFS_LIC_ISFREE(licp, i)) {
6114 lidp = XFS_LIC_SLOT(licp, i);
6115 lip = lidp->lid_item;
6116 if (lip->li_type != XFS_LI_BUF)
6119 bip = (xfs_buf_log_item_t *)lip;
6122 if (XFS_BUF_ADDR(lbp) == bno) {
6124 break; /* Found it */
6127 licp = licp->lic_next;
6135 xfs_bmbt_block_t *block,
6141 xfs_bmbt_ptr_t *pp, *thispa; /* pointer to block address */
6142 xfs_bmbt_key_t *prevp, *keyp;
6144 ASSERT(be16_to_cpu(block->bb_level) > 0);
6147 for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6148 dmxr = mp->m_bmap_dmxr[0];
6151 keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6153 keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
6154 xfs_bmbt, block, i, dmxr);
6158 xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6163 * Compare the block numbers to see if there are dups.
6167 pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6169 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6170 xfs_bmbt, block, i, dmxr);
6172 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6174 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6176 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6177 xfs_bmbt, block, j, dmxr);
6179 if (INT_GET(*thispa, ARCH_CONVERT) ==
6180 INT_GET(*pp, ARCH_CONVERT)) {
6181 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6183 INT_GET(*thispa, ARCH_CONVERT));
6184 panic("%s: ptrs are equal in node\n",
6192 * Check that the extents for the inode ip are in the right order in all
6197 xfs_bmap_check_leaf_extents(
6198 xfs_btree_cur_t *cur, /* btree cursor or null */
6199 xfs_inode_t *ip, /* incore inode pointer */
6200 int whichfork) /* data or attr fork */
6202 xfs_bmbt_block_t *block; /* current btree block */
6203 xfs_fsblock_t bno; /* block # of "block" */
6204 xfs_buf_t *bp; /* buffer for "block" */
6205 int error; /* error return value */
6206 xfs_extnum_t i=0, j; /* index into the extents list */
6207 xfs_ifork_t *ifp; /* fork structure */
6208 int level; /* btree level, for checking */
6209 xfs_mount_t *mp; /* file system mount structure */
6210 xfs_bmbt_ptr_t *pp; /* pointer to block address */
6211 xfs_bmbt_rec_t *ep; /* pointer to current extent */
6212 xfs_bmbt_rec_t *lastp; /* pointer to previous extent */
6213 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
6216 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6222 ifp = XFS_IFORK_PTR(ip, whichfork);
6223 block = ifp->if_broot;
6225 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6227 level = be16_to_cpu(block->bb_level);
6229 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6230 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6231 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6232 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6233 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6234 bno = INT_GET(*pp, ARCH_CONVERT);
6236 * Go down the tree until leaf level is reached, following the first
6237 * pointer (leftmost) at each level.
6239 while (level-- > 0) {
6240 /* See if buf is in cur first */
6241 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6247 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6248 XFS_BMAP_BTREE_REF)))
6250 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6251 XFS_WANT_CORRUPTED_GOTO(
6252 XFS_BMAP_SANITY_CHECK(mp, block, level),
6258 * Check this block for basic sanity (increasing keys and
6259 * no duplicate blocks).
6262 xfs_check_block(block, mp, 0, 0);
6263 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6264 1, mp->m_bmap_dmxr[1]);
6265 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6266 bno = INT_GET(*pp, ARCH_CONVERT);
6269 xfs_trans_brelse(NULL, bp);
6274 * Here with bp and block set to the leftmost leaf node in the tree.
6279 * Loop over all leaf nodes checking that all extents are in the right order.
6283 xfs_fsblock_t nextbno;
6284 xfs_extnum_t num_recs;
6287 num_recs = be16_to_cpu(block->bb_numrecs);
6290 * Read-ahead the next leaf block, if any.
6293 nextbno = be64_to_cpu(block->bb_rightsib);
6296 * Check all the extents to make sure they are OK.
6297 * If we had a previous block, the last entry should
6298 * conform with the first entry in this one.
6301 ep = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6302 block, 1, mp->m_bmap_dmxr[0]);
6303 for (j = 1; j < num_recs; j++) {
6304 nextp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6305 block, j + 1, mp->m_bmap_dmxr[0]);
6307 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6308 (void *)lastp, (void *)ep);
6310 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6319 xfs_trans_brelse(NULL, bp);
6323 * If we've reached the end, stop.
6325 if (bno == NULLFSBLOCK)
6328 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6334 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6335 XFS_BMAP_BTREE_REF)))
6337 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6341 xfs_trans_brelse(NULL, bp);
6346 cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6348 xfs_trans_brelse(NULL, bp);
6350 cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6352 panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6358 * Count fsblocks of the given fork.
6361 xfs_bmap_count_blocks(
6362 xfs_trans_t *tp, /* transaction pointer */
6363 xfs_inode_t *ip, /* incore inode */
6364 int whichfork, /* data or attr fork */
6365 int *count) /* out: count of blocks */
6367 xfs_bmbt_block_t *block; /* current btree block */
6368 xfs_fsblock_t bno; /* block # of "block" */
6369 xfs_ifork_t *ifp; /* fork structure */
6370 int level; /* btree level, for checking */
6371 xfs_mount_t *mp; /* file system mount structure */
6372 xfs_bmbt_ptr_t *pp; /* pointer to block address */
6376 ifp = XFS_IFORK_PTR(ip, whichfork);
6377 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6378 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6379 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6381 XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6382 XFS_ERRLEVEL_LOW, mp);
6383 return XFS_ERROR(EFSCORRUPTED);
6389 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6391 block = ifp->if_broot;
6392 level = be16_to_cpu(block->bb_level);
6394 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6395 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6396 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6397 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6398 bno = INT_GET(*pp, ARCH_CONVERT);
6400 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6401 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6403 return XFS_ERROR(EFSCORRUPTED);
6410 * Recursively walks each level of a btree
6411 * to count total fsblocks is use.
6414 xfs_bmap_count_tree(
6415 xfs_mount_t *mp, /* file system mount point */
6416 xfs_trans_t *tp, /* transaction pointer */
6417 xfs_ifork_t *ifp, /* inode fork pointer */
6418 xfs_fsblock_t blockno, /* file system block number */
6419 int levelin, /* level in btree */
6420 int *count) /* Count of blocks */
6423 xfs_buf_t *bp, *nbp;
6424 int level = levelin;
6426 xfs_fsblock_t bno = blockno;
6427 xfs_fsblock_t nextbno;
6428 xfs_bmbt_block_t *block, *nextblock;
6431 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6434 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6437 /* Not at node above leafs, count this level of nodes */
6438 nextbno = be64_to_cpu(block->bb_rightsib);
6439 while (nextbno != NULLFSBLOCK) {
6440 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6441 0, &nbp, XFS_BMAP_BTREE_REF)))
6444 nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6445 nextbno = be64_to_cpu(nextblock->bb_rightsib);
6446 xfs_trans_brelse(tp, nbp);
6449 /* Dive to the next level */
6450 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6451 xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6452 bno = INT_GET(*pp, ARCH_CONVERT);
6453 if (unlikely((error =
6454 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6455 xfs_trans_brelse(tp, bp);
6456 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6457 XFS_ERRLEVEL_LOW, mp);
6458 return XFS_ERROR(EFSCORRUPTED);
6460 xfs_trans_brelse(tp, bp);
6462 /* count all level 1 nodes and their leaves */
6464 nextbno = be64_to_cpu(block->bb_rightsib);
6465 numrecs = be16_to_cpu(block->bb_numrecs);
6466 if (unlikely(xfs_bmap_disk_count_leaves(ifp, mp,
6467 0, block, numrecs, count) < 0)) {
6468 xfs_trans_brelse(tp, bp);
6469 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6470 XFS_ERRLEVEL_LOW, mp);
6471 return XFS_ERROR(EFSCORRUPTED);
6473 xfs_trans_brelse(tp, bp);
6474 if (nextbno == NULLFSBLOCK)
6477 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6478 XFS_BMAP_BTREE_REF)))
6481 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6488 * Count leaf blocks given a range of extent records.
6491 xfs_bmap_count_leaves(
6498 xfs_bmbt_rec_t *frp;
6500 for (b = 0; b < numrecs; b++) {
6501 frp = xfs_iext_get_ext(ifp, idx + b);
6502 *count += xfs_bmbt_get_blockcount(frp);
6508 * Count leaf blocks given a range of extent records originally
6512 xfs_bmap_disk_count_leaves(
6516 xfs_bmbt_block_t *block,
6521 xfs_bmbt_rec_t *frp;
6523 for (b = 1; b <= numrecs; b++) {
6524 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6525 xfs_bmbt, block, idx + b, mp->m_bmap_dmxr[0]);
6526 *count += xfs_bmbt_disk_get_blockcount(frp);